Detection method and system for false triggering of automobile tail door, storage medium and intelligent automobile

Abstract

The invention relates to the technical field of vehicle control, in particular to a detection method and system for false triggering of an automobile tail door, a storage medium and an intelligent automobile. The method comprises the steps that capacitance value data of each capacitance sensor at multiple time points in the foot kicking action process of a tester are obtained to serve as a time sequence, space sequence data are constructed by combining speed data of the multiple time points, and the size of the time sequence is set according to the speed data; acquiring space sequence data corresponding to the kicking actions of the plurality of testers, labeling to generate training data, and training the convolutional neural network model; when a capacitance signal is detected, setting the size of a time sequence according to the current speed data, storing the change of the capacitance value data of each capacitance sensor until the final time point of the time sequence is reached, obtaining a to-be-identified time sequence, constructing to-be-identified space sequence data in combination with the current speed data, and storing the to-be-identified space sequence data; and inputting the image into a convolutional neural network model for identification. According to the invention, the detection robustness and adaptability can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle control, and in particular to a method, system, storage medium and intelligent vehicle for detecting false triggering of a tailgate of an automobile. Background Art

[0002] With the rapid development of automotive electronic technology, people's requirements for the comfort and convenience of cars are getting higher and higher. As a major component of a car, the tailgate greatly guarantees the internal safety of the car. It is particularly important to open or close the tailgate safely and effectively. The existing method of opening the tailgate by kicking has been widely used in many vehicles due to its convenience. This function is realized through a capacitive sensor at the bottom of the vehicle. When you approach the vehicle and kick it lightly under the rear bumper, the tailgate will automatically open or close. This design is especially suitable for operation when both hands are inconvenient, such as when returning from shopping or carrying large items, which greatly improves the convenience of daily use.

[0003] In the above implementation, the existing specific approach is to use a neural network to determine whether the change in capacitance value within a certain period of time is a human kicking action. However, the disadvantage of this method is that it has a weak ability to distinguish non-target actions (such as slight touches, environmental vibrations, etc.), a low recognition rate for false triggers, and poor adaptability and robustness to the recognition of trigger operations of different people.

[0004] Therefore, the existing detection method for false triggering of the automobile tailgate has a low recognition rate for false triggering, and has poor robustness and adaptability. Summary of the invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a method, system, storage medium and smart car for detecting false triggering of a car tailgate, so as to solve the problem that the existing detection method for false triggering of a car tailgate in the prior art has a low recognition rate for false triggering, poor robustness and poor adaptability.

[0006] The present invention discloses a method for detecting false triggering of a car tailgate. A speed sensor and a plurality of capacitance sensors with different directions are arranged on the car tailgate. The method for detecting false triggering of the car tailgate comprises:

[0007] Acquire the capacitance value data of each capacitance sensor at multiple time points during the tester's kicking action as a time series, and construct a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data;

[0008] Acquire spatial sequence data corresponding to the kicking actions of multiple testers;

[0009] Annotate each set of spatial sequence data to generate training data, train the convolutional neural network model using the training data, and obtain a pre-trained convolutional neural network model;

[0010] When a capacitance signal is detected, the size of the time series is set according to the speed data currently obtained by the speed sensor, and the change of the capacitance value data of each capacitance sensor is stored until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the spatial sequence data to be identified in combination with the current speed data, and input the spatial sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

[0011] Optionally, the method for detecting false triggering of the automobile tailgate further includes:

[0012] Get the temperature and humidity data of the environment;

[0013] Performing a denoising operation on the capacitance value data according to the temperature data and the humidity data;

[0014] The capacitance value data after the denoising operation is used as a time series.

[0015] Optionally, the performing a denoising operation on the capacitance value data according to the temperature data and the humidity data comprises the following steps:

[0016] Determining whether the temperature data and the humidity data are abnormal;

[0017] If the temperature data or the humidity data is abnormal, the difference between the acquired capacitance value data and a preset reference value is calculated as the capacitance value data after denoising.

[0018] Optionally, the step of labeling each set of spatial sequence data to generate training data comprises the following steps:

[0019] Mark according to the actual triggering situation corresponding to each set of spatial sequence data;

[0020] Multiple sets of spatial sequence data are combined to form a three-dimensional array as training data. The first dimension of the three-dimensional array represents the sweeping process of the feet of different testers, the second dimension represents different time points, and the third dimension is capacitance data, speed data and acceleration data.

[0021] Optionally, the convolutional neural network model is established by the following steps:

[0022] Build a convolutional neural network;

[0023] Performing a first convolution on the convolutional neural network to obtain a first tensor, the size of the first convolution kernel is 64*64, the number of the first convolution kernel is 32, and the activation function of the first convolution is a ReLU function;

[0024] Performing a first pooling on the first tensor to obtain a corresponding second tensor, the size of the first pooling kernel is 8*8, the step size of the first pooling is 2, and the activation function of the first pooling is a ReLU function;

[0025] A second convolution is performed on the second tensor shown to obtain a corresponding third tensor, the size of the second convolution kernel is 64*64, the number of the second convolution kernel is 16, and the activation function of the second convolution is the ReLU function;

[0026] Perform a second pooling on the third tensor to obtain a corresponding fourth tensor, the size of the second pooling kernel is 8*8, the step size of the second pooling is 2, and the activation function of the second pooling is the ReLU function;

[0027] Performing a first full connection on the fourth tensor to obtain a corresponding fifth tensor, the number of nodes in the first fully connected layer is 32, the activation function of the first full connection is a ReLU function, and the size of the fifth tensor is 1*32;

[0028] The fifth tensor is fully connected for the second time to obtain the corresponding sixth tensor, the activation function of the second full connection is the Sigmoid function, the size of the sixth tensor is 1*2, and the two elements in the sixth tensor respectively represent the probabilities corresponding to false triggering and non-false triggering, thereby obtaining a convolutional neural network model.

[0029] Optionally, the formula of the ReLU function is:

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

[0031] The formula of the Sigmoid function is:

[0032]

[0033] The formula of the loss function of the convolutional neural network is:

[0034]

[0035] Among them, x is the input value and y is the true value.

[0036] Optionally, the method for detecting false triggering of a car tailgate further comprises the following steps:

[0037] The sensing mode of the capacitive sensor is set to a mixed mode, and the capacitive sensor obtains a digital count value as capacitance value data according to a conversion formula corresponding to the mixed mode, and the mixed mode includes a mutual capacitance mode and a self-capacitance mode;

[0038] The conversion formula corresponding to the self-contained mode is as follows:

[0039]

[0040] The conversion formula corresponding to the mutual capacitance mode is as follows:

[0041]

[0042] in,

[0043] Maxcount = N sub *TxClk Div

[0044] C S =C P +C F

[0045] Maxcount is the maximum number of digital counts converted from capacitance values, N sub is the number of sub-conversions, TxClk Div SnsClk is the Tx clock division rate. Div is the sensing clock division frequency, C S is the inductive capacitance, C ref is the reference capacitor, C P is the parasitic capacitance, C F is the human foot capacitance; C comp CompClk is the compensation capacitor Div To compensate for the clock division rate.

[0046] The present invention also discloses a detection system for false triggering of a car tailgate, wherein a speed sensor and a plurality of capacitance sensors with different directions are arranged on the car tailgate, and the detection system for false triggering of the car tailgate comprises:

[0047] a conversion module, used to obtain the capacitance value data of each of the capacitance sensors at multiple time points during the tester's kicking action as a time series, and to construct a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data;

[0048] An acquisition module, used for acquiring spatial sequence data corresponding to the kicking actions of multiple testers;

[0049] A training module, used to annotate each set of spatial sequence data to generate training data, train the convolutional neural network model through the training data, and obtain a pre-trained convolutional neural network model;

[0050] The detection module is used to set the size of the time series according to the speed data currently obtained by the speed sensor when a capacitance signal is detected, and store the change of the capacitance value data of each capacitance sensor until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the space sequence data to be identified in combination with the current speed data, and input the space sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

[0051] The present invention also discloses an intelligent car, comprising 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 described above.

[0052] The present invention also discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method described above.

[0053] Compared with the prior art, the beneficial effect of the detection method for false triggering of the automobile tailgate provided by the embodiment of the present invention is that: by obtaining the capacitance value data of multiple capacitance sensors in different directions at multiple time points during the tester's kicking action as a time series, combining the speed data of the kicking action to construct the entire spatial sequence data of the kicking action process, and using the spatial sequence data of multiple testers as training data for the convolutional neural network model, and when detecting whether it is a false trigger, the spatial sequence data constructed by combining the capacitance data and the speed data is also used as the input of the convolutional neural network model for detection and identification. It can more accurately identify whether the behavior of triggering the capacitance signal is a false trigger, has a higher false trigger recognition rate, and has higher robustness. In addition, the size of the time series is set according to the speed data, and different testers have different time series. Therefore, the pre-trained convolutional neural network model has better adaptability to different people. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The scheme of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which:

[0055] Figure 1 1 is a flow chart of a method for detecting false triggering of a car tailgate provided by an embodiment of the present invention;

[0056] Figure 2 is a schematic diagram of the structure of a convolutional neural network provided by an embodiment of the present invention;

[0057] Figure 3 It is a module block diagram of a detection system for false triggering of a car tailgate provided by an embodiment of the present invention;

[0058] Figure 4 is a module block diagram of a smart car provided by an embodiment of the present invention;

[0059] Figure 5 It is a module block diagram of a computer storage medium provided by an embodiment of the present invention.

[0060] The reference numerals in the figures are:

[0061] 100. Detection system for false triggering of the tailgate of a car;

[0062] 110, conversion module; 120, acquisition module; 130, training module; 140, detection module; 150, setting module;

[0063] 200. Smart cars;

[0064] 210, processor; 220, memory;

[0065] 300. Computer readable storage medium;

[0066] 310. Computer program. DETAILED DESCRIPTION

[0067] 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.

[0068] An embodiment of the present invention provides a method for detecting false triggering of a car tailgate, which is used for opening the car tailgate to identify whether the operation behavior on the car tailgate is a false trigger. When it is identified as a normal triggering behavior, the car tailgate is opened; when it is identified as a false trigger, no operation is performed.

[0069] In an embodiment of the present application, a speed sensor and multiple capacitance sensors facing different directions are provided on the tailgate of the automobile. The capacitance sensors facing different directions can obtain the capacitance of the human foot corresponding to the kicking action from multiple directions and convert it into corresponding capacitance value data.

[0070] like Figure 1 As shown, the detection method of false triggering of the tailgate of a car includes:

[0071] S110, acquiring capacitance value data of each capacitance sensor at multiple time points during the tester's kicking action as a time series, and constructing a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data;

[0072] S120, obtaining spatial sequence data corresponding to the kicking actions of multiple testers;

[0073] S130, annotating each set of spatial sequence data to generate training data, training the convolutional neural network model through the training data, and obtaining a pre-trained convolutional neural network model;

[0074] S140. When a capacitance signal is detected, the size of the time series is set according to the speed data currently obtained by the speed sensor, and the change of the capacitance value data of each capacitance sensor is stored until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the spatial sequence data to be identified in combination with the current speed data, and input the spatial sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

[0075] The present application obtains the capacitance value data of multiple time points during the tester's kicking action of multiple capacitance sensors in different directions as a time series, and combines the speed data of the kicking action to construct the entire spatial sequence data during the kicking action. The spatial sequence data of multiple testers are used as the training data of the convolutional neural network model, and when detecting whether it is falsely triggered, the spatial sequence data constructed by combining the capacitance data and the speed data is also used as the input of the convolutional neural network model for detection. The convolutional neural network model is good at processing data with spatial structure, and can more accurately identify whether the behavior of triggering the capacitance signal is a false trigger. It has a high false trigger recognition rate and high robustness, which improves the user experience and enhances the security of the system. In addition, the size of the time series is set according to the speed data, and different testers have different time series. Therefore, the pre-trained convolutional neural network model has better adaptability to different people. The detection method of the embodiment of the present application has a false trigger recognition rate of up to 98.72%, which is higher than other detection methods.

[0076] In step S110 , the tester's kicking action process refers to the process from the start of the kicking action to the end of the kicking action.

[0077] Time series (or dynamic series) refers to a series formed by arranging the values ​​of the same statistical indicator in the order of their occurrence. The time series formed in step S110 includes the capacitance value data of the capacitance sensors at the corresponding time points at multiple time points and different orientations. The size of the time series is not fixed, and the duration of the time series is not fixed. Different people have different time series data, and the size of the time series is set according to the speed data and acceleration data of the kicking action, that is, the speed of the kicking action. For example, if a person kicks the capacitance sensor at a fast speed, the time series is reduced, and the data of a few time points are collected less. If the speed of the kick is slow, the time series is increased, so that the capacitance sensor and the speed sensor can collect more data at a few time points, so that the pre-trained convolutional neural network model can be better adapted to each person, and the acceleration data can be further processed by the speed data measured by the speed sensor. In the process of training the convolutional neural network model, people of different ages, genders, weights, and exercise habits can be invited to participate in data collection so that the model can better adapt to different situations.

[0078] Capacitive sensors in different orientations obtain capacitance data corresponding to the kicking action from multiple directions, providing more comprehensive data from multiple angles, which helps to more accurately capture the characteristics and patterns of the kicking action. Even if a sensor is interfered with or fails, other sensors can still provide valid data to ensure the normal operation of the system and increase the robustness of detection. Combining data from multiple sensors can enhance the recognition and detection accuracy of the trigger signal. Capacitive sensors in different orientations have different capacitance values ​​at each time point in the time series.

[0079] In a specific implementation scenario, three capacitive sensors are set. For the same spatial position, one capacitive sensor faces directly downward to sense the spatial position, one capacitive sensor faces right to sense the spatial position, and the remaining capacitive sensor faces left to sense the spatial position.

[0080] The steps of using the capacitance value data of multiple capacitance sensors as a time series and combining the speed data of the speed sensor to construct the spatial series data include:

[0081] Obtaining capacitance value data of each capacitance sensor at each time point in the time series;

[0082] Acquire speed data collected by the speed sensor at multiple time points, and obtain acceleration data based on the speed data;

[0083] The capacitance value data, velocity data and acceleration data are constructed into spatial sequence data, which includes a one-dimensional array, wherein the first dimension data of the two-dimensional array are different time points, and the second dimension data are the capacitance value data, velocity data and acceleration data.

[0084] In the embodiment of the present application, the type of the capacitive sensor is proximity sensing, and the method for detecting false triggering of the tailgate of a car further includes the following steps:

[0085] The sensing mode of the capacitive sensor is set to a mixed mode, and the capacitive sensor obtains a digital count value as capacitance value data according to a conversion formula corresponding to the mixed mode, and the mixed mode includes a mutual capacitance mode and a self-capacitance mode;

[0086] The conversion formula corresponding to the self-contained mode is as follows:

[0087]

[0088] The conversion formula corresponding to the mutual capacitance mode is as follows:

[0089]

[0090] in,

[0091] Maxcount = N sub *TxClk Div

[0092] C S =C P +C F

[0093] Maxcount is the maximum number of digital counts converted from capacitance values, N sub is the number of sub-conversions, TxClk Div To send the clock division rate, SnsClk Div is the sensing clock division frequency, C S is the inductive capacitance, C ref is the reference capacitor, C P is the parasitic capacitance, C F is the human foot capacitance; C comp CompClk is the compensation capacitor Div To compensate for the clock division rate.

[0094] According to the capacitance increment of the human foot passing over the capacitance sensor, the circuit inside the capacitance sensor converts the capacitance value of the human foot into a corresponding digital count value and outputs it as capacitance value data.

[0095] Specifically, in the above conversion formula, 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 the internal circuit of the capacitance sensor, a complete conversion cycle is usually divided into multiple sub-conversion steps, each of which processes and converts the input capacitance to obtain a digital output result.

[0096] Transmit clock frequency TxClk Div Refers to the frequency division of the clock signal used to control the operation of the transmitter in the capacitive sensor system. This frequency determines the rate at which the sensor collects and transmits data, and has a certain impact on the response speed and accuracy of the system.

[0097] Sensing clock frequency SnsClk Div , usually refers to the minimum time change capability that the sensor can perceive, which is used to control the clock frequency of the sensing capacitance measurement and affects the accuracy and speed of the measurement. Div , the time interval of the inductive capacitance measurement can be controlled, thereby affecting the accuracy and sensitivity of the measurement.

[0098] Parasitic capacitance C P It is the unintentional capacitance in the circuit of the capacitive sensor due to structure, wiring or other factors.

[0099] Compensation capacitor C comp Used to correct errors or deviations that may exist in the measurement process of capacitance sensors. Compensation capacitors are usually used to adjust the sensitivity, stability and accuracy of capacitance sensors to improve the performance of the sensor.

[0100] Compensation clock division rate CompClk Div The role of the compensation capacitor C comp The frequency division process is performed to adjust the influence of the compensation capacitor, which in turn affects the calculation. The compensation capacitor C can be corrected by the frequency division operation of the compensation divider. comp The error introduced ensures accurate data in dual CDAC mode.

[0101] Reference capacitor C ref Used as a reference value to help measure changes in target capacitance. Specifically, it provides a known fixed value for calibrating and comparing changes in target capacitance. ref By comparing the capacitance of the target object with that of the touch sensor, the change in capacitance of the target object can be accurately measured, thereby realizing the detection and recognition of the external environment or touch input.

[0102] Human foot capacitance C F When a person's foot approaches the sensor surface, the capacitance around the foot will affect the capacitance value in the sensor circuit, thereby generating a signal that can be detected. F It can be measured by a capacitive sensor installed on the tailgate of the car.

[0103] Due to its structural characteristics, the mutual capacitance mode has good stability to environmental changes such as humidity and temperature, and has strong recognition ability for different materials. The self-capacitance mode can be flexibly adjusted to adapt to different environmental conditions and is more flexible in sensitivity adjustment. Setting the sensing mode of the capacitive sensor to the mixed mode, that is, using self-capacitance sensing and mutual capacitance sensing at the same time, and making full use of their respective advantages, can effectively reduce false triggering caused by environmental factors, achieve better detection results, improve system reliability, and improve the overall performance of the system.

[0104] In order to reduce the influence of environmental factors on the detection results of the capacitive sensor, in the embodiment of the present application, the detection method of the false triggering of the tailgate of the automobile further includes:

[0105] Get the temperature and humidity data of the environment;

[0106] Perform denoising operation on capacitance value data according to temperature data and humidity data;

[0107] The capacitance value data after the denoising operation is used as a time series.

[0108] A temperature sensor can be used to collect the temperature data of the environment, a humidity sensor can be used to collect the humidity data of the environment, or a temperature and humidity sensor can be used to collect the temperature and humidity data of the environment at the same time. The temperature and humidity sensor mostly uses an integrated temperature and humidity probe as a temperature measuring element to collect temperature and humidity signals. After voltage stabilization and filtering, operational amplification, nonlinear correction, V / I conversion, constant current and reverse protection and other circuit processing, it is converted into a current signal or voltage signal output that is linearly related to temperature and humidity, that is, the temperature data and humidity data can be output.

[0109] When the temperature of the environment is abnormal, such as the temperature is higher than the preset value or drops to below zero, or the capacitive sensor is covered with water, the data tested by the capacitive sensor will produce errors, which will lead to false triggering. De-noising the capacitance value data based on temperature data and humidity data can effectively solve the changes in capacitance signals caused by changes in the surrounding environment. The capacitance value data after denoising is used as a time series and as the data basis for constructing spatial sequence data, providing a more accurate data basis for spatial sequence data and improving the recognition rate of false triggers by the trained convolutional neural network model.

[0110] Specifically, the capacitance value data is denoised according to the temperature data and the humidity data, including the following steps:

[0111] Determine whether the temperature data and humidity data are abnormal;

[0112] If the temperature data or the humidity data is abnormal, the difference between the acquired capacitance value data and the preset reference value is calculated as the capacitance value data after denoising.

[0113] The preset reference value is the influence of temperature or humidity on the capacitance value data of the capacitance sensor tested in the experimental environment. Temperature and humidity have corresponding preset reference values. When the temperature data or humidity data is abnormal, the difference between the actual capacitance value data and the preset reference value is obtained as the actual capacitance value data, eliminating the interference caused by changes in environmental conditions, and providing input data with high accuracy and reliability for the training of the convolutional neural network model.

[0114] Specifically, if only the temperature of the environment is in an abnormal state, such as the temperature is higher than the preset value or drops to below zero, the difference between the obtained capacitance value data and the preset reference value corresponding to the temperature abnormality is calculated as the denoised capacitance value data. If only the humidity is abnormal, indicating that there is water attached to the capacitance sensor, the difference between the obtained capacitance value data and the preset reference value corresponding to the humidity abnormality is calculated as the denoised capacitance value data. If both the temperature and humidity are abnormal, the difference is superimposed and calculated as the denoised capacitance value data.

[0115] In order to provide more reliable basic data, digital filters such as low-pass filters and band-pass filters can also be used to filter out noise within a specific frequency range. You can select the appropriate filter type according to specific needs, process the capacitance value data, remove environmental noise, interference signals or other unnecessary frequency components, extract the required signal information, and improve the quality and reliability of the data.

[0116] In step S130, each set of spatial sequence data is labeled to generate training data, including the following steps:

[0117] Mark according to the actual triggering situation corresponding to each set of spatial sequence data;

[0118] Multiple sets of spatial sequence data are merged to form a three-dimensional array as training data. The first dimension of the three-dimensional array represents the sweeping process of the feet of different testers, the second dimension represents different time points, and the third dimension is the capacitance value data, speed data and acceleration data.

[0119] By labeling multiple sets of spatial sequence data, the convolutional neural network model can better identify false triggering behaviors. Processing multiple sets of spatial sequence data into three-dimensional arrays as training data, each dimension contains information of the same attributes, which makes it easier for the convolutional neural network model to better understand and learn the complex relationship between data, and to better train the convolutional neural network model on the data, thereby improving the false triggering recognition rate.

[0120] In a specific implementation scenario, the spatial sequence data can be labeled with "0" for no trigger and "1" for trigger, providing accurate labels for the convolutional neural network model so that the model can learn how to identify triggers.

[0121] In addition, multiple groups of spatial sequence data corresponding to a batch of different testers scanning the capacitive sensor multiple times can be collected as test sets. These test sets are also labeled according to the actual triggering conditions. The multiple groups of spatial sequence data are merged to form a three-dimensional array as training data. The first dimension of the three-dimensional array represents the scanning process of the feet of different testers, the second dimension represents different time points, and the third dimension is the capacitance value data, speed data and acceleration data. The obtained data is input into the convolutional neural network model to test and evaluate the performance and generalization ability of the model.

[0122] Furthermore, in order to help the convolutional neural network model better understand the characteristics of the data and improve the performance and generalization ability of the model, all training data are normalized before training the convolutional neural network model so that all data are distributed in the range of 0 to 1. This is proposed for the convenience of data processing. Mapping the data to the range of 0 to 1 helps accelerate the convergence speed of the neural network. When the range of feature values ​​is very different, the neural network may require more iterations to converge to the optimal solution. Through normalization, the value ranges of each feature can be made similar, thereby reducing training time; data normalization can also increase the stability of the model. The neural network is more sensitive to changes in input data. If the data is not normalized, the model may be greatly affected by slight changes in the data; normalization can help reduce the risk of overfitting of the model. Overfitting usually occurs when the model overfits the training data. Normalization can help the model better generalize to unseen data; and through normalization, the design of the neural network model can be simplified. Normalization can make the network structure more concise and reduce the number of hyperparameters that need to be adjusted.

[0123] Specifically, the minimum-maximum normalization method can be used to linearly scale the data to a specific range, find the maximum and minimum values ​​in the training data, and process each training data using the following data normalization formula:

[0124]

[0125] Among them, X i Each piece of data in the data set; X min The minimum data in the data set; X max The maximum value in the data set.

[0126] The convolutional neural network model is built through the following steps:

[0127] Build a convolutional neural network;

[0128] The first tensor is obtained by performing the first convolution on the convolutional neural network. The size of the first convolution kernel is 64*64, the number of the first convolution kernel is 32, and the activation function of the first convolution is the ReLU function.

[0129] The first tensor is pooled for the first time to obtain the corresponding second tensor. The size of the first pooling kernel is 8*8, the step size of the first pooling is 2, and the activation function of the first pooling is the ReLU function.

[0130] The second tensor shown is convolved for the second time to obtain the corresponding third tensor. The size of the second convolution kernel is 64*64, the number of the second convolution kernel is 16, and the activation function of the second convolution is the ReLU function.

[0131] The third tensor is pooled for the second time to the corresponding fourth tensor. The size of the second pooling kernel is 8*8, the step size of the second pooling is 2, and the activation function of the second pooling is the ReLU function.

[0132] The fourth tensor is fully connected for the first time to obtain the corresponding fifth tensor. The number of nodes in the first fully connected layer is 32. The activation function of the first fully connected layer is the ReLU function. The size of the fifth tensor is 1*32.

[0133] The fifth tensor is fully connected for the second time to obtain the corresponding sixth tensor. The activation function of the second full connection is the Sigmoid function. The size of the sixth tensor is 1*2. The two elements in the sixth tensor represent the probabilities corresponding to false triggering and non-false triggering, respectively, to obtain the convolutional neural network model. The structural diagram of the convolutional neural network model is shown in the figure. Figure 2 shown.

[0134] The formula of the ReLU function is:

[0135] Relu(x)=max(0,x)

[0136] Where x is the input value. For input x, if x is greater than 0, then output x, otherwise output 0.

[0137] The formula for the Sigmoid function is:

[0138]

[0139] Map the input value to between 0 and 1 for the sixth tensor to perform probability prediction for the binary classification problem.

[0140] The formula for the loss function of the convolutional neural network is:

[0141]

[0142] The loss function uses the cross entropy loss function to measure the difference between the model output and the actual label, y i is the actual label, x i is the predicted value of the model output, and n is the number of samples. The cross entropy loss function is commonly used in binary classification problems.

[0143] In the process of establishing the above convolutional neural network model, through the stacking of convolutional layers and pooling layers, the convolutional neural network model can effectively extract features from the input data, help identify patterns and regularities in spatial sequence data, and help the model better understand the data; the fully connected layer can integrate and classify the features extracted by convolution, thereby helping the model to make the final decision. The first fully connected layer can further improve the representation ability of the model by designing the number of nodes as 32 and the activation function as ReLU.

[0144] After building the convolutional neural network model, set the training rounds and data batches. The training round is the number of times the neural network iterates on the entire training set, and the data batch refers to the number of samples used for each model update. A larger batch size can speed up training but may take up more memory. In general, the choice of batch size depends on the size of the dataset and the computing resources.

[0145] Input the training data into the convolutional neural network model for training and testing. Perform multiple rounds of training based on the training data, and continuously adjust the weights and biases of the convolutional neural network model through the back propagation algorithm so that the model can better fit the data set. During the training process, save the parameters and state of the convolutional neural network model regularly to prevent accidental data loss. Saving the convolutional neural network model can be achieved by saving the model's weights, structure, and optimizer state to a file. Then, deploy the trained convolutional network to the MCU to detect the triggering of the car's tailgate.

[0146] In step 140, when a human foot or other non-target object sweeps over the capacitive sensor, it will cause a change in the capacitive signal. The pre-trained convolutional neural network model can identify the behavior that causes the capacitive signal change and determine whether the behavior is a false trigger. If it is a false trigger, no operation is performed. If it is a trigger, the tailgate of the car is controlled to open.

[0147] like Figure 3 As shown, an embodiment of the present invention further provides a detection system 100 for false triggering of a car tailgate. A speed sensor and a plurality of capacitive sensors with different directions are arranged on the car tailgate, and the detection system 100 for false triggering of the car tailgate comprises:

[0148] The conversion module 110 is used to obtain the capacitance value data of each capacitance sensor at multiple time points during the tester's kicking action as a time series, and to construct a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data;

[0149] An acquisition module 120 is used to acquire spatial sequence data corresponding to the kicking actions of multiple testers;

[0150] A training module 130 is used to annotate each set of spatial sequence data to generate training data, train the convolutional neural network model through the training data, and obtain a pre-trained convolutional neural network model;

[0151] The detection module 140 is used to set the size of the time series according to the speed data currently obtained by the speed sensor when a capacitance signal is detected, and store the change of the capacitance value data of each capacitance sensor until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the space sequence data to be identified in combination with the current speed data, and input the space sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

[0152] The conversion module 110 is also used to obtain temperature data and humidity data of the environment; perform denoising operation on the capacitance value data according to the temperature data and humidity data; and use the capacitance value data after denoising operation as a time series. The conversion module 110 performs denoising operation on the capacitance value data according to the temperature data and humidity data, including the following steps: determining whether the temperature data and humidity data are abnormal; if the temperature data or humidity data is abnormal, calculating the difference between the obtained capacitance value data and the preset reference value as the capacitance value data after denoising.

[0153] The training module 130 is also specifically used to mark according to the actual trigger situation corresponding to each group of spatial sequence data; merge multiple groups of spatial sequence data to form a three-dimensional array as training data, the first dimension of the three-dimensional array represents the process of the feet of different testers sweeping, the second dimension represents different time points, and the third dimension is capacitance value data, speed data and acceleration data.

[0154] The automobile tailgate false triggering detection system 100 also includes a setting module 150, which is used to set the sensing mode of the capacitive sensor to a mixed mode. The capacitive sensor obtains a digital count value as capacitance value data according to a conversion formula corresponding to the mixed mode. The mixed mode includes a mutual capacitance mode and a self-capacitance mode.

[0155] The conversion formula corresponding to the self-contained mode is as follows:

[0156]

[0157] The conversion formula corresponding to the mutual capacitance mode is as follows:

[0158]

[0159] in,

[0160] Maxcount = N sub *TxClk Div

[0161] C S =C P +C F

[0162] Maxcount is the maximum number of digital counts converted from capacitance values, N sub is the number of sub-conversions, TxClk Div SnsClk is the Tx clock division rate. Div is the sensing clock division frequency, C S is the inductive capacitance, C ref is the reference capacitor, C P is the parasitic capacitance, C F is the human foot capacitance; C comp CompClk is the compensation capacitor Div To compensate for the clock division rate.

[0163] Based on the above, it can be known that the detection system for false triggering of the automobile tailgate in the embodiment of the present application obtains the capacitance value data of multiple capacitance sensors in different directions at multiple time points during the tester's kicking action as a time series, and constructs the entire spatial sequence data of the kicking action process in combination with the speed data of the kicking action. The spatial sequence data of multiple testers are used as training data for the convolutional neural network model, and when detecting whether it is a false trigger, the spatial sequence data constructed by combining the capacitance data and the speed data is also used as the input of the convolutional neural network model for detection. It can more accurately identify whether the behavior of triggering the capacitance signal is a false trigger, has a higher false trigger recognition rate, and has higher robustness. The detection system can operate more safely and stably. In addition, the size of the time series is set according to the speed data, and different testers have different time series. Therefore, the pre-trained convolutional neural network model has better adaptability to different people.

[0164] like Figure 4 As shown, an embodiment of the present invention further provides a smart car 200. The smart car 200 includes a processor 210 and a memory 220. The processor 210 is coupled to the memory 220. The memory 220 stores a computer program 310, and the processor 210 executes the computer program 310 when working to implement the above method. The detailed steps can be referred to above, and will not be repeated here.

[0165] like Figure 5 As shown, an embodiment of the present invention further provides a computer-readable storage medium 300. The computer-readable storage medium 300 stores at least one computer program 310, which is used to be executed by the processor 210 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 300 can be a storage chip, a hard disk, or a mobile hard disk or a USB flash drive, an optical disk, or other readable and writable storage tools in the terminal, or a server, etc.

[0166] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments 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 the present 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).

[0167] 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.

[0168] 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 false triggering of a car tailgate, characterized in that: The automobile tailgate is provided with a speed sensor and a plurality of capacitance sensors facing different directions, and the detection method of false triggering of the automobile tailgate comprises: Acquire the capacitance value data of each capacitance sensor at multiple time points during the tester's kicking action as a time series, and construct a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data; Acquire spatial sequence data corresponding to the kicking actions of multiple testers; Annotate each set of spatial sequence data to generate training data, train the convolutional neural network model using the training data, and obtain a pre-trained convolutional neural network model; When a capacitance signal is detected, the size of the time series is set according to the speed data currently obtained by the speed sensor, and the change of the capacitance value data of each capacitance sensor is stored until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the spatial sequence data to be identified in combination with the current speed data, and input the spatial sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

2. The method for detecting false triggering of a car tailgate according to claim 1, characterized in that: The method for detecting false triggering of the automobile tailgate further comprises: Get the temperature and humidity data of the environment; Performing a denoising operation on the capacitance value data according to the temperature data and the humidity data; The capacitance value data after the denoising operation is used as a time series.

3. The method for detecting false triggering of a car tailgate according to claim 2, characterized in that: The denoising operation on the capacitance value data according to the temperature data and the humidity data comprises the following steps: Determining whether the temperature data and the humidity data are abnormal; If the temperature data or the humidity data is abnormal, the difference between the acquired capacitance value data and a preset reference value is calculated as the capacitance value data after denoising.

4. The method for detecting false triggering of a car tailgate according to claim 1, characterized in that: The step of labeling each set of spatial sequence data to generate training data includes the following steps: Mark according to the actual triggering situation corresponding to each set of spatial sequence data; Multiple sets of spatial sequence data are combined to form a three-dimensional array as training data. The first dimension of the three-dimensional array represents the sweeping process of the feet of different testers, the second dimension represents different time points, and the third dimension is capacitance data, speed data and acceleration data.

5. The method for detecting false triggering of a car tailgate according to any one of claims 1 to 4, characterized in that: The convolutional neural network model is established by the following steps: Build a convolutional neural network; Performing a first convolution on the convolutional neural network to obtain a first tensor, the size of the first convolution kernel is 64*64, the number of the first convolution kernel is 32, and the activation function of the first convolution is a ReLU function; Performing a first pooling on the first tensor to obtain a corresponding second tensor, the size of the first pooling kernel is 8*8, the step size of the first pooling is 2, and the activation function of the first pooling is a ReLU function; A second convolution is performed on the second tensor shown to obtain a corresponding third tensor, the size of the second convolution kernel is 64*64, the number of the second convolution kernel is 16, and the activation function of the second convolution is the ReLU function; Perform a second pooling on the third tensor to obtain a corresponding fourth tensor, the size of the second pooling kernel is 8*8, the step size of the second pooling is 2, and the activation function of the second pooling is the ReLU function; Performing a first full connection on the fourth tensor to obtain a corresponding fifth tensor, the number of nodes in the first fully connected layer is 32, the activation function of the first full connection is a ReLU function, and the size of the fifth tensor is 1*32; The fifth tensor is fully connected for the second time to obtain the corresponding sixth tensor, the activation function of the second full connection is the Sigmoid function, the size of the sixth tensor is 1*2, and the two elements in the sixth tensor respectively represent the probabilities corresponding to false triggering and non-false triggering, thereby obtaining a convolutional neural network model.

6. The method for detecting false triggering of a car tailgate according to claim 5, characterized in that: The formula of the ReLU function is: Relu(x)=max(0,x) The formula of the Sigmoid function is: The formula of the loss function of the convolutional neural network is: Among them, x is the input value and y is the true value.

7. The method for detecting false triggering of a car tailgate according to any one of claims 1 to 4, characterized in that: The method for detecting false triggering of the automobile tailgate further comprises the following steps: The sensing mode of the capacitive sensor is set to a mixed mode, and the capacitive sensor obtains a digital count value as capacitance value data according to a conversion formula corresponding to the mixed mode, and the mixed mode includes a mutual capacitance mode and a self-capacitance mode; The conversion formula corresponding to the self-contained mode is as follows: The conversion formula corresponding to the mutual capacitance mode is as follows: in, Maxcount=N sub *TxClk Div C S =C P +C F Maxcount is the maximum number of digital counts converted from capacitance values, N sub is the number of sub-conversions, TxClk Div SnsClk is the Tx clock division rate. Div is the sensing clock division frequency, C S is the inductive capacitance, C ref is the reference capacitor, C P is the parasitic capacitance, C F is the human foot capacitance; C comp CompClk is the compensation capacitor Div To compensate for the clock division rate.

8. A detection system for false triggering of a car tailgate, characterized in that: The automobile tailgate is provided with a speed sensor and a plurality of capacitance sensors with different directions, and the automobile tailgate false triggering detection system comprises: a conversion module, used to obtain the capacitance value data of each of the capacitance sensors at multiple time points during the tester's kicking action as a time series, and to construct a set of spatial series data in combination with the speed data collected by the speed sensor at multiple time points, wherein the size of the time series is set according to the speed data; An acquisition module, used for acquiring spatial sequence data corresponding to the kicking actions of multiple testers; A training module, used to annotate each set of spatial sequence data to generate training data, train the convolutional neural network model through the training data, and obtain a pre-trained convolutional neural network model; The detection module is used to set the size of the time series according to the speed data currently obtained by the speed sensor when a capacitance signal is detected, and store the change of the capacitance value data of each capacitance sensor until the final time point of the time series is reached, so as to obtain the time series to be identified, and construct the space sequence data to be identified in combination with the current speed data, and input the space sequence data to be identified into the pre-trained convolutional neural network model for identification, so as to detect whether the behavior of triggering the capacitance signal is a false trigger.

9. 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.

10. 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.

Images