Steering wheel hand-leaving false triggering detection method and system, storage medium and intelligent automobile

Abstract

The invention relates to the technical field of vehicle detection, in particular to a steering wheel hand-leaving false triggering detection method and system, a storage medium and an intelligent automobile, and the method comprises the steps: collecting multiple groups of capacitance change data generated in the process from the beginning of approaching a steering wheel to the touch of the steering wheel by a capacitance sensor; cleaning and marking the acquired capacitance change data; training the labeled capacitance change data by using a neural network to obtain a classification model, and deploying the classification model to a single-chip microcomputer; acquiring current capacitance change data, and inputting the current capacitance change data into the classification model to identify whether the current behavior is false triggering; according to the scheme, the capacitance change data generated in the process from approaching the steering wheel to touching the steering wheel by a large number of different people in different angles and different touch modes are used as the model training data to carry out model training and realize automatic detection, a capacitance threshold value does not need to be manually set, errors caused by manual operation are reduced, and the detection efficiency is improved. And the misjudgment risk is reduced, so that the safety of vehicle driving is improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle detection technology, and in particular to a method, system, storage medium and intelligent vehicle for detecting hand-off accidental triggering of a steering wheel. Background Art

[0002] With the acceleration of urbanization and the continuous increase in the number of cars, the automatic driving of cars is becoming more and more common. However, with the development of automatic driving, safety issues also arise. According to regulations, when driving automatically, people cannot leave the steering wheel.

[0003] Most of the steering wheel hand-off detection on the market currently determines whether a person's hand touches the steering wheel by setting a capacitance threshold. When the hand touches the sensor on the steering wheel, the sensor senses an increase or decrease in the capacitance signal value and compares it with the average capacitance signal value when it is not touched, thereby determining whether there is a hand touch. However, different people have different finger capacitances, and the palm area covered by touching the steering wheel and the amount of force applied will cause the capacitance signal received by the sensor to be different. If the threshold is set too small, it will be detected as a touch when the palm has not yet touched the steering wheel; and if the threshold is set too large, it may cause the palm to touch the steering wheel but still be judged as not touched. If you simply set a touch threshold, there will be a risk of false touches, and the recognized gestures will not be accurate.

[0004] Therefore, it is of vital importance to those skilled in the art to design a steering wheel hands-off mis-triggering detection method, system, storage medium and smart car that can effectively prevent hands-off mis-triggering. Summary of the invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a steering wheel hand-off mis-triggering detection method, system, storage medium and smart car that can effectively prevent hand-off mis-triggering, so as to solve the problems of mis-touch risk and inaccurate gesture recognition in the prior art.

[0006] The present invention discloses a method for detecting a hands-off steering wheel mis-triggering, which comprises the following steps:

[0007] Capacitive sensors are used to collect data on capacitance changes generated by multiple groups of hands from the time they start approaching the steering wheel to the time they touch the steering wheel;

[0008] Cleaning the acquired capacitance change data, and marking each group of capacitance change data after cleaning;

[0009] The neural network is used to train the labeled capacitance change data to obtain a classification model, and the trained classification model is deployed on the microcontroller;

[0010] When a capacitance signal is detected, the change of the capacitance signal starts to be stored until a preset time is reached to end the storage to obtain the current capacitance change data, and the current capacitance change data is input into the classification model to identify whether the current behavior is a false trigger.

[0011] Optionally, the “collecting capacitance change data generated by multiple groups of human hands from the beginning of approaching the steering wheel to touching the steering wheel using a capacitive sensor” comprises the following steps:

[0012] Each time a human hand approaches the steering wheel and touches the steering wheel, a set of capacitance change values ​​is collected, and multiple sets are collected at a preset collection interval;

[0013] The mode of the capacitance sensor is set to a mutual capacitance mode, and each capacitance value in each group of capacitance change values ​​is converted into corresponding capacitance data according to a preset conversion formula to form corresponding capacitance change data.

[0014] Optionally, the conversion formula of the capacitive sensor in single CDAC mode is:

[0015]

[0016] The conversion formula of the capacitive sensor in dual CDAC mode is:

[0017]

[0018] in,

[0019] Maxcount = N sub *TxCLK Div

[0020] Rawdata is the touch data, Maxcount is the maximum touch capacitance data collected by the capacitance sensor; N sub is the sub-conversion number of the capacitive sensor, TxCLK Div is the Tx clock division frequency of the capacitive sensor, CP is the parasitic capacitance value of the capacitive sensor, CF is the finger capacitance value collected by the capacitive sensor, Cref is the reference capacitance value of the capacitive sensor, and CompClkDiv is the compensation CDAD division frequency of the capacitive sensor.

[0021] Optionally, the “cleaning the acquired capacitance change data and marking each group of capacitance change data after cleaning” includes the following steps:

[0022] The preset marking rule is “1” for accidental touch and “0” for no accidental touch.

[0023] The invalid data and abnormal data in the capacitance change data are filtered out by a digital filter;

[0024] Each set of capacitance change data after cleaning is labeled according to a preset labeling rule to form a label set.

[0025] Optionally, the “using a neural network to train the labeled capacitance change data to obtain a classification model, and deploying the trained classification model on a single chip computer” includes the following steps:

[0026] Select part of the data from all the capacitance change data as a training set, and another part of the data as a test set;

[0027] Import the training set and test set in Python, and load the capacitance change data and corresponding label set in the training set;

[0028] The classification model is trained multiple times using the capacitance change data and corresponding label set in the training set, and the trained classification model is deployed to the microcontroller.

[0029] Optionally, the classification model is a 6-layer fully connected neural network, which includes an input layer, 4 hidden layers and an output layer, the input layer has 128 nodes, the 4 hidden layers have 32, 32, 16, and 16 nodes respectively, the output layer has 2 nodes, the hidden layer is activated by a Relu function, and the output layer is activated by a Sigmoid function, wherein the Relu activation function is:

[0030] Relu(x)=max(0,x)

[0031] The Sigmoid activation function is:

[0032]

[0033] The loss function of the fully connected neural network is:

[0034]

[0035] Optionally, the step of “obtaining current capacitance change data, and inputting the current capacitance change data into a classification model to identify whether the current behavior is a false trigger” includes the following steps:

[0036] Input the current capacitance change data into the classification model for classification, and select the result with the largest probability as the prediction result;

[0037] According to the prediction result, if it is a false trigger, no processing will be performed; if it is not a false trigger, continue to determine whether it is currently a touch, if it is a false trigger, no processing will be performed, if it is not a false trigger, a warning will be issued.

[0038] In order to solve the problems existing in the prior art, the present invention further provides a steering wheel hands-off mis-triggering detection system, the solution of which is to include:

[0039] A data acquisition module, used to collect capacitance change data;

[0040] Data processing and annotation module, used for cleaning and classification annotation of capacitance change data;

[0041] A model training module, used to train a classification model using the labeled capacitance change data and deploy the trained classification model to the controller;

[0042] A detection module is used to collect current capacitance change data and input the current capacitance change data into a classification model to identify whether the current behavior is a false trigger;

[0043] Warning module, used to issue a warning when the steering wheel is not touched.

[0044] In order to solve the problems existing in the prior art, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor executes the steps of any of the methods described above.

[0045] In order to solve the problems existing in the prior art, the present invention also provides a smart car, which includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any of the methods described above.

[0046] Compared with the prior art, the beneficial effect of the steering wheel hands-off false trigger detection method provided by the embodiment of the present invention is: by collecting a large number of capacitance change data generated in the process from approaching the steering wheel to touching the steering wheel at different angles and in different touch methods, and using a neural network to train a large amount of capacitance change data to obtain a classification model, and using the classification model and the current capacitance change data to determine whether the current hands-off trigger is a false trigger, and issuing a warning when it is not a false trigger. This scheme uses the capacitance change data generated in the process from approaching the steering wheel to touching the steering wheel as a training set for model training, thereby realizing automatic detection of hands-off false triggering, without the need to manually set the capacitance threshold, reducing the error caused by human operation, reducing the risk of misjudgment, and thus improving the safety of vehicle driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The scheme of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which:

[0048] Figure 1 The process of the steering wheel hands-off mis-triggering detection method provided by the embodiment of the present invention is as follows: Figure 1 ;

[0049] Figure 2 The process of the steering wheel hands-off mis-triggering detection method provided by the embodiment of the present invention is as follows: Figure 2 ;

[0050] Figure 3 The process of the steering wheel hands-off mis-triggering detection method provided by the embodiment of the present invention is as follows: Figure 3 ;

[0051] Figure 4 is a system block diagram of a steering wheel hands-off mis-triggering detection system provided by an embodiment of the present invention;

[0052] Figure 5 It is a system block diagram of a smart car provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. Now, in conjunction with the accompanying drawings, a preferred embodiment of the present invention is described in detail.

[0054] like Figure 1 As shown, the present invention provides a specific embodiment of a method for detecting a hands-off steering wheel mis-triggering.

[0055] A method for detecting the accidental triggering of the steering wheel leaving the hands, reference Figure 1 , including the following steps:

[0056] S1, using a capacitive sensor to collect capacitance change data generated by multiple groups of hands from the beginning of approaching the steering wheel to touching the steering wheel;

[0057] S2, cleaning the acquired capacitance change data, and marking each group of capacitance change data after cleaning;

[0058] S3, using a neural network to train the labeled capacitance change data to obtain a classification model, and deploying the trained classification model to the single chip microcomputer;

[0059] S4. When a capacitance signal is detected, the change of the capacitance signal starts to be stored until a preset time is reached and the storage ends to obtain the current capacitance change data, and the current capacitance change data is input into the classification model to identify whether the current behavior is a false trigger.

[0060] Specifically, multiple testers with different body characteristics touch the steering wheel multiple times, and all capacitance values ​​generated in each process from approaching the steering wheel to touching the steering wheel are collected, and all capacitance values ​​generated in each process from approaching the steering wheel to touching the steering wheel are grouped together, and multiple groups are collected at a preset collection interval; in this embodiment, the collection interval for each process can be 10ms, and a total of 128 capacitance values ​​are collected in each process to represent the capacitance signal of the entire process from initial approach to complete touching.

[0061] Further, the mode of the capacitance sensor is set to a mutual capacitance mode, and each capacitance value in each group of capacitance change values ​​is converted into corresponding capacitance data according to a preset conversion formula to form corresponding capacitance change data.

[0062] Specifically, the mode of the capacitive sensor is set to the mutual capacitance mode, which is based on the parasitic capacitance between the two electrodes. When a finger approaches or touches the electrode, a new capacitance is formed between the electrodes, which affects the original mutual capacitance value, so as to determine whether a touch occurs by the change in the capacitance value; when the capacitive sensor is in the single CDAC mode (there is only a reference capacitance DAC in the capacitive sensor, and the compensation capacitance DAC is in the off state), each capacitance data is calculated according to the following conversion formula:

[0063]

[0064] Among them, Rawdata is the capacitance data, N sub is the sub-conversion number of the capacitive sensor, TxCLK Div is the Tx clock division frequency of the capacitive sensor, CP is the parasitic capacitance value of the capacitive sensor, and CF is the finger capacitance value collected by the capacitive sensor.

[0065] When the capacitance sensor is in dual CDAC mode (both the reference capacitance DAC and the compensation capacitance DAC are enabled in the capacitance sensor), each capacitance data is calculated according to the following conversion formula:

[0066]

[0067] in,

[0068] Maxcount = N sub *TxCLK Div ;

[0069] Rawdata is the capacitance data, Maxcount is the maximum capacitance data collected by the capacitance sensor; N sub is the sub-conversion number of the capacitive sensor, TxCLK Divis the Tx clock division frequency of the capacitive sensor, CP is the parasitic capacitance value of the capacitive sensor, CF is the finger capacitance value collected by the capacitive sensor, Ccomp is the compensation capacitance value of the capacitive sensor, CompClkDiv is the compensation CDAD division frequency of the capacitive sensor, and Cref is the reference capacitance value of the capacitive sensor.

[0070] In CDAC, the number of sub-conversions N sub Refers to the number of sub-conversion steps required to convert the input capacitance value into a digital signal within a complete conversion cycle. In CDAC, a complete conversion cycle is usually divided into multiple sub-conversion steps, each of which performs certain processing and conversion on the input capacitance to obtain a digital output result.

[0071] The parasitic capacitance CP is an unintentional capacitance existing in the circuit of the capacitive sensor due to structure, wiring or other factors.

[0072] Finger capacitance CF is when a finger approaches the surface of the capacitive sensor. The capacitance around the finger will affect the capacitance value in the capacitive sensor circuit, thereby generating a detectable signal. The finger capacitance CF can be measured and obtained by the capacitive sensors in different partitions of the steering wheel. It can be any of the touch capacitance data, current capacitance data, and reference capacitance data.

[0073] The compensation capacitor Ccomp is used to correct the error or deviation that may exist in the measurement process of the capacitive sensor. The compensation capacitor is usually used to adjust the sensitivity, stability and accuracy of the capacitive sensor to improve the performance of the sensor.

[0074] The role of the compensation divider frequency CompClkDiv here is to divide the compensation capacitor Ccomp to adjust the influence of the compensation capacitor, thereby affecting the calculation. Through the frequency division operation of the compensation divider, the error introduced by the compensation capacitor Ccomp can be corrected to ensure accurate data in the dual CDAC mode.

[0075] The reference capacitance value Cref is used as a reference value to help measure the change of the target capacitance. Specifically, it provides a known fixed value for calibrating and comparing the change of the target capacitance. By comparing with the reference capacitance Cref, the change of the target capacitance can be accurately measured, thereby realizing the detection and recognition of the external environment or touch input.

[0076] Tx clock division frequency TxCLK DivThe clock frequency used to control the sensing capacitance measurement affects the accuracy and speed of the measurement. It is usually used to divide the system clock signal to generate the sensing clock signal for measurement. By adjusting the division ratio of the sensing clock divider, the time interval of the sensing capacitance measurement can be controlled, thereby affecting the accuracy and sensitivity of the measurement.

[0077] All capacitance values ​​collected in each process can be converted into corresponding capacitance data through the above conversion formula, and all capacitance data in each process can be summarized into a group to form capacitance change data for each process from approaching the steering wheel to touching the steering wheel.

[0078] In one embodiment, reference Figure 2 “Cleaning the acquired capacitance change data and marking each group of capacitance change data after cleaning” includes the following steps:

[0079] S21, preset labeling rules, "1" for accidental touch, "0" for no accidental touch;

[0080] S22, filtering out invalid data and abnormal data in the capacitance change data through a digital filter;

[0081] S23 , labeling each group of capacitance change data after cleaning according to a preset labeling rule to form a label set.

[0082] Specifically, a labeling rule is preset first, and the labeling rule is used to classify and label each result. The labeling rule may be to use "1" to represent a false touch and "0" to represent no false touch. The obtained capacitance change data is cleaned, and a digital filter (such as a low-pass filter, a band-pass filter, etc.) can be used to filter out noise within a specific frequency range to remove invalid data and abnormal data, thereby forming a data set for subsequent training.

[0083] Furthermore, the cleaned capacitance change data is classified according to preset labeling rules and corresponding classification labels are added to form a label set for subsequent training; the label set can provide the model with accurate false trigger judgment basis, so that the model can automatically make false trigger judgments based on the current capacitance change data.

[0084] In one embodiment, reference Figure 3 “Using a neural network to train the labeled capacitance change data to obtain a classification model, and deploying the trained classification model to the microcontroller” includes the following steps:

[0085] S31, selecting part of the data from all the capacitance change data as a training set, and another part of the data as a test set;

[0086] S32, importing the training set and the test set in Python, and loading the capacitance change data and the corresponding label set in the training set;

[0087] S33, performing multiple rounds of training on the classification model using the capacitance change data in the training set and the corresponding label set, and deploying the trained classification model to the single chip microcomputer.

[0088] Specifically, most of the labeled capacitance change data is selected as a training set, specifically 95% of the capacitance change data, for training the classification model; and the remaining small part of the data is selected as a test set, specifically 5% of the capacitance change data, for evaluating the accuracy of the classification model.

[0089] Among them, in order to process the data more quickly and conveniently, the capacitance change data in the training set and the test set can be normalized before training. The normalization process generally includes minimum-maximum scaling, Z-score standardization, normal distribution normalization, etc.; according to the data distribution and model requirements in this embodiment, the minimum-maximum scaling method is selected for normalization.

[0090] Minimum-maximum normalization is to scale the data to a fixed range through linear transformation, usually [0,1] range or [-1,1] range, calculate the minimum min and maximum max of each data, and apply the minimum-maximum scaling formula to each value x of each data. Specifically, the normalization can be obtained according to the following formula, and the difference data is implied to the range of [0,1]:

[0091]

[0092] Among them, X i For each data in the data set; X min is the minimum data in the data set; X max The maximum value in the data set.

[0093] Through minimum-maximum normalization data processing, not only can the original distribution of the data be retained and the data only be scaled to a fixed range, but the impact of different dimensions between different data can also be eliminated, making the data more comparable; when training a neural network model, normalizing the data can speed up the convergence of the model to reduce training time and resource consumption, help the model better learn the relationship between the distribution and features of the data, and improve the performance and generalization ability of the model.

[0094] To use a neural network to train the normalized data, you can first set the training rounds and data batches, then import the training set and test set into the neural network model built in Python, and load the capacitance change data and corresponding label set in the training set for training; a training round is the number of times the entire training set is iteratively trained on the neural network. In each training round, the neural network receives the entire data set and updates the weights through forward propagation, loss calculation, back propagation and other steps. The appropriate number of training rounds can be determined based on the training effect of the observed model; in each training round, the data set is divided into multiple batches for training, and each data batch contains a certain number of sample data. The neural network processes each data batch one by one, calculates the corresponding loss and updates the weights.

[0095] The classification model is trained using annotated datasets to minimize prediction errors through repeated iterations. Finally, the trained classification model is converted into a format suitable for running on a microcontroller by compression, quantization, or conversion to a format suitable for the target hardware, and the converted classification model is deployed to the target microcontroller.

[0096] In one embodiment, the classification model is a 6-layer fully connected neural network, which includes an input layer, 4 hidden layers and an output layer. The input layer has 128 nodes, the 4 hidden layers have 32, 32, 16, and 16 nodes respectively, the output layer has 2 nodes, the hidden layer is activated by a Relu function, and the output layer is activated by a Sigmoid function.

[0097] Specifically, the classification model is a 6-layer fully connected neural network. The input layer of the classification model includes 128 nodes to represent the capacitance data of each time from the finger approaching to completely touching the steering wheel; the hidden layer of the classification model has 4 layers, the first hidden layer includes 32 nodes, the second hidden layer includes 32 nodes, the third hidden layer includes 16 nodes, and the fourth hidden layer includes 16 nodes; the output layer of the classification model includes 2 nodes.

[0098] Furthermore, each hidden layer of the classification model is activated by the Relu function. The Relu function is a nonlinear activation function that can help the neural network learn nonlinear relationships to improve the network's expressiveness. The Relu function is simple and efficient to calculate, and there is no problem of gradient disappearance. When the input is a negative number, the output of the Relu function is 0, which also brings a certain sparsity to the model, which helps the generalization ability of the model.

[0099] The Relu function is:

[0100] Relu(x)=max(0,x)

[0101] Among them, if the input x is greater than or equal to 0, the output of the Relu activation function is x, if the output x is less than 0, the output of the Relu activation function is 0.

[0102] Furthermore, the output layer of the classification model is activated by the Sigmoid function. The Sigmoid function can map real numbers to a range between 0 and 1. The Sigmoid function has good stability for small changes in the input and can effectively reduce the instability in model training. The Sigmoid function is a continuous and differentiable function with good smoothness, which makes it easier for the gradient descent algorithm to converge during training. Since the output of the Sigmoid function can be interpreted as probability, using the Sigmoid function to activate the output layer can provide an effective probabilistic interpretation of the results.

[0103] The Sigmoid activation function is:

[0104]

[0105] Among them, each element x is each capacitance data, and the Sigmoid function converts each element x into a value between 0 and 1 to represent the probability of the classification.

[0106] The fully connected neural network uses a binary cross entropy loss function for loss calculation. Assuming that for each sample, the true label is y (0 or 1), the output of the model is x (ranging from 0 to 1), the loss function of the fully connected neural network is:

[0107]

[0108] Among them, n is the number of samples, y is the true label of the sample (0 or 1), and x is the model's predicted value for the sample. When the true label is 1, the loss function will punish the model output of 0, so that the model learns in the right direction. When the true label is 0, the loss function will punish the model output of 1. By minimizing the binary cross entropy loss function, the model can better fit the training data and improve the classification accuracy.

[0109] In one of the embodiments, after the current capacitance change data is input into the classification model for classification, the result with the largest probability is selected as the prediction result; if the prediction result is a false trigger, no processing is performed; if it is not a false trigger, it continues to determine whether it is currently a touch, if it is a false trigger, no processing is performed, and if it is not a false trigger, a warning is issued.

[0110] like Figure 4 As shown, the present invention provides a specific embodiment of a steering wheel hands-off mis-triggering detection system.

[0111] A steering wheel hands-off mis-triggering detection system is used to implement the steering wheel hands-off mis-triggering detection method as described in any one of the above items, referring to Figure 4 The steering wheel hands-off accidental trigger detection system includes: a data acquisition module 100, a data processing and annotation module 200, a model training module 300, a detection module 400 and a warning module 500.

[0112] Specifically, the data acquisition module 100 is used to collect multiple groups of capacitance change values ​​generated from the finger approaching the steering wheel to touching the steering wheel, and convert the capacitance change value in each group into corresponding capacitance change data according to a preset conversion formula.

[0113] Furthermore, the data processing and annotation module 200 is used to clean the obtained capacitance change data. Specifically, a digital filter can be configured in the data processing and annotation module 200, and a digital filter (such as a low-pass filter, a band-pass filter, etc.) can be used to filter out noise within a specific frequency range to remove invalid data and abnormal data to form a data set for subsequent training.

[0114] The data processing and labeling module 200 is also used to classify the capacitance change data after cleaning according to preset labeling rules and add corresponding classification labels to form a label set for subsequent training; wherein the labeling rule may be to indicate false touch with "1" and to indicate no false touch with "0"; the data processing and labeling module 200 classifies the capacitance change data after cleaning and adds corresponding classification labels according to the labeling rule of "indicate false touch with "1" and indicate no false touch with "0"" to form a label set for subsequent training; the label set can provide the model with accurate false trigger judgment basis, so that the model can automatically make false trigger judgment based on the current capacitance change data.

[0115] The model training module 300 is used to train the classification model using labeled capacitance change data to minimize the prediction error through repeated iterations; finally, the trained classification model is converted into a format suitable for running on a microcontroller through operations such as compression, quantization or conversion into a format suitable for the target hardware, and the converted classification model is deployed to the target microcontroller.

[0116] Among them, the classification model is a 6-layer fully connected neural network. The input layer of the classification model includes 128 nodes to represent the capacitance data of each time from the finger approaching to completely touching the steering wheel; the hidden layer of the classification model has 4 layers, the first hidden layer includes 32 nodes, the second hidden layer includes 32 nodes, the third hidden layer includes 16 nodes, and the fourth hidden layer includes 16 nodes; the output layer of the classification model includes 2 nodes.

[0117] Furthermore, each hidden layer of the classification model is activated by the Relu function. The Relu function is a nonlinear activation function that can help the neural network learn nonlinear relationships to improve the network's expressiveness. The Relu function is simple and efficient to calculate, and there is no problem of gradient disappearance. When the input is a negative number, the output of the Relu function is 0, which also brings a certain sparsity to the model, which helps the generalization ability of the model.

[0118] The Relu function is:

[0119] Relu(x)=max(0,x)

[0120] Among them, if the input x is greater than or equal to 0, the output of the Relu activation function is x, if the output x is less than 0, the output of the Relu activation function is 0.

[0121] Furthermore, the output layer of the classification model is activated by the Sigmoid function. The Sigmoid function can map real numbers to a range between 0 and 1. The Sigmoid function has good stability for small changes in the input and can effectively reduce the instability in model training. The Sigmoid function is a continuous and differentiable function with good smoothness, which makes it easier for the gradient descent algorithm to converge during training. Since the output of the Sigmoid function can be interpreted as probability, using the Sigmoid function to activate the output layer can provide an effective probabilistic interpretation of the results.

[0122] The Sigmoid activation function is:

[0123]

[0124] Among them, each element x is each capacitance data, and the Sigmoid function converts each element x into a value between 0 and 1 to represent the probability of the classification.

[0125] The fully connected neural network uses a binary cross entropy loss function for loss calculation. Assuming that for each sample, the true label is y (0 or 1), the output of the model is x (ranging from 0 to 1), the loss function of the fully connected neural network is:

[0126]

[0127] Among them, n is the number of samples, y is the true label of the sample (0 or 1), and x is the model's predicted value for the sample. When the true label is 1, the loss function will punish the model output of 0, so that the model learns in the right direction. When the true label is 0, the loss function will punish the model output of 1. By minimizing the binary cross entropy loss function, the model can better fit the training data and improve the classification accuracy.

[0128] The detection module 400 is used to collect current capacitance change data and input the capacitance change data into a classification model to obtain a detection result.

[0129] The warning module 500 is used to issue a warning. When the current capacitance change data is input into the classification model for classification, the result with the largest probability is selected as the prediction result. If the prediction result is a false trigger, no processing is performed. If it is not a false trigger, it continues to determine whether the current touch is a touch. If it is a false trigger, no processing is performed. If it is not a false trigger, the warning module 500 is controlled to issue a warning.

[0130] The present invention also provides a specific embodiment of a computer-readable storage medium.

[0131] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of the steering wheel hands-off mis-triggering detection method as described in any one of the above items.

[0132] Among them, at least one computer program is stored in the computer-readable storage medium, and the computer program is used to be executed by the processor to implement the above method. The detailed steps can be referred to above and will not be repeated here. In one embodiment, the computer-readable storage medium can be a storage chip, a hard disk, or a mobile hard disk or a USB flash drive, a CD, or other readable and writable storage tools in the terminal, or a server, etc.

[0133] like Figure 5 As shown, the present invention also provides a specific embodiment of a smart car.

[0134] A smart car, reference Figure 5 The smart car includes a memory 10 and a processor 20. The memory 10 stores a computer program. When the computer program is executed by the processor 20, the processor 20 executes the steps of the steering wheel hand-off mis-triggering detection method as described in any one of the above items.

[0135] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database 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. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0136] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. For those skilled in the art, the technical solutions described in the above embodiments can be modified, or some of the technical features therein can be replaced by equivalents; and all these modifications and replacements should fall within the protection scope of the claims attached to the present invention.

Claims

1. A method for detecting a hands-off steering wheel accidental triggering, characterized in that: The following steps are involved: Capacitive sensors are used to collect data on capacitance changes generated by multiple groups of hands from the time they start approaching the steering wheel to the time they touch the steering wheel; Cleaning the acquired capacitance change data, and marking each group of capacitance change data after cleaning; The neural network is used to train the labeled capacitance change data to obtain a classification model, and the trained classification model is deployed on the microcontroller; When a capacitance signal is detected, the change of the capacitance signal starts to be stored until a preset time is reached to end the storage to obtain the current capacitance change data, and the current capacitance change data is input into the classification model to identify whether the current behavior is a false trigger.

2. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 1, characterized in that: The “collecting capacitance change data generated by multiple groups of hands from approaching the steering wheel to touching the steering wheel using a capacitance sensor” comprises the following steps: Each time a human hand approaches the steering wheel and touches the steering wheel, a set of capacitance change values ​​is collected, and multiple sets are collected at a preset collection interval; The mode of the capacitance sensor is set to a mutual capacitance mode, and each capacitance value in each group of capacitance change values ​​is converted into corresponding capacitance data according to a preset conversion formula to form corresponding capacitance change data.

3. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 2, characterized in that: The conversion formula of the capacitive sensor in single CDAC mode is: The conversion formula of the capacitive sensor in dual CDAC mode is: in, Maxcount=N sub *TxCLK Div Rawdata is the touch data, Maxcount is the maximum touch capacitance data collected by the capacitance sensor; N sub is the sub-conversion number of the capacitive sensor, TxCLK Div is the Tx clock division frequency of the capacitive sensor, CP is the parasitic capacitance value of the capacitive sensor, CF is the finger capacitance value collected by the capacitive sensor, Cref is the reference capacitance value of the capacitive sensor, and CompClkDiv is the compensation CDAD division frequency of the capacitive sensor.

4. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 1, characterized in that: The “cleaning the acquired capacitance change data and marking each group of capacitance change data after cleaning” includes the following steps: Preset labeling rules, "1" for false touch, "0" for no false touch; The invalid data and abnormal data in the capacitance change data are filtered out by a digital filter; Each set of capacitance change data after cleaning is labeled according to a preset labeling rule to form a label set.

5. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 4, characterized in that: The “using a neural network to train the labeled capacitance change data to obtain a classification model, and deploying the trained classification model on a single chip microcomputer” includes the following steps: Select part of the data from all the capacitance change data as a training set, and another part of the data as a test set; Import the training set and test set in Python, and load the capacitance change data and corresponding label set in the training set; The classification model is trained multiple times using the capacitance change data and corresponding label set in the training set, and the trained classification model is deployed to the microcontroller.

6. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 1, characterized in that: The classification model is a 6-layer fully connected neural network, which includes an input layer, 4 hidden layers and an output layer. The input layer has 128 nodes, the 4 hidden layers have 32, 32, 16, and 16 nodes respectively, the output layer has 2 nodes, the hidden layer is activated by a Relu function, and the output layer is activated by a Sigmoid function, wherein the Relu activation function is: Relu(x)=max(0,x) The Sigmoid activation function is: The loss function of the fully connected neural network is:

7. The method for detecting the accidental triggering of the hands-off steering wheel according to claim 6, characterized in that: The "obtaining current capacitance change data, and inputting the current capacitance change data into a classification model to identify whether the current behavior is a false trigger" includes the following steps: Input the current capacitance change data into the classification model for classification, and select the result with the largest probability as the prediction result; According to the prediction result, if it is a false trigger, no processing will be performed; if it is not a false trigger, continue to determine whether it is currently a touch, if it is a false trigger, no processing will be performed, if it is not a false trigger, a warning will be issued.

8. A steering wheel hands-off mis-triggering detection system, characterized in that: include: A data acquisition module, used to collect capacitance change data; Data processing and annotation module, used for cleaning and classification annotation of capacitance change data; A model training module is used to train a classification model using the labeled capacitance change data and deploy the trained classification model to the controller; A detection module is used to collect current capacitance change data and input the current capacitance change data into a classification model to identify whether the current behavior is a false trigger; Warning module, used to issue a warning when the steering wheel is not touched.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

10. A smart car, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

Images