Vehicle state detection method, medium and equipment

Abstract

The invention relates to the technical field of vehicle positioning, in particular to a vehicle state detection method, medium and equipment, and the method comprises the following steps: obtaining the data of an IMU sensor installed on a vehicle; performing low-pass filtering processing on the IMU data; calculating a multi-scale absolute deviation integral statistic of the long and short windows based on the processed IMU data, and constructing a statistic time sequence feature vector; and constructing a TCN time convolution network based on the time sequence feature vector, training the TCN time convolution network according to the label data to obtain a model, and judging the motion state by using the trained model. According to the method, the motion state rule is learned from the multi-scale absolute deviation integral statistics by using the TCN time convolutional network model, the vehicle motion state in a continuous low-speed straight line or frequent starting scene under GNSS long-term lock losing can be effectively identified, and the method has very high vibration resistance and is suitable for both fuel vehicles and electric vehicles.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle positioning, in particular to a vehicle state detection method, medium and equipment. BACKGROUND

[0002] In the extreme scenario that the GNSS (Global Navigation Satellite System) of the vehicle is long-term invalid and the vehicle is in a complex motion mode, the traditional algorithm cannot effectively identify the motion state of the vehicle, thereby causing the navigation and location service of the vehicle to be paralyzed, the emergency brake to be mis-triggered / failure, the lane keeping function to be disorderly, and the like, affecting the safety of vehicle driving. The traditional algorithm specifically has the following problems: (1) Noise interference problem under low-speed motion: when the vehicle travels at a constant low speed (<0.5 m / s), the noise power spectral density of the MEMS (Micro-Electro-Mechanical System) accelerometer is similar to the magnitude of the motion acceleration, which causes the detection algorithm (MAG (Magnitudes of Acceleration and Gyroscope, three-axis acceleration and angular velocity amplitude detection method), ARE (Angular Rate Energy, angular rate energy detection method)) based on a fixed threshold to be insufficient in signal-to-noise ratio and to produce high-frequency mis-triggering; (2) Model mismatch problem of quasi-static creeping working condition: in the frequent start-stop scene, the vehicle is often in a micro-motion state of 0.1-0.5 m / s, and the detection model under the traditional static assumption cannot represent the gradual change characteristics of the time-frequency domain features, resulting in continuous misjudgment at the dynamic-static state transition boundary; (3) Lack of anti-vibration performance: due to the significant difference in engine vibration energy levels between electric vehicles and fuel vehicles, the method based on a fixed threshold (such as MAG and ARE) only has certain effect on electric vehicles with small vibration, but the detection performance sharply decreases on fuel vehicles with large vibration. On the contrary, the frequency domain analysis method based on Fast Fourier Transform (FFT) has certain effect on fuel vehicles with large vibration, but it cannot detect the static state on electric vehicles at all.

[0003] In summary, there is an urgent need for a method that can not only effectively identify the motion state of the vehicle in the extreme scenario that the GNSS of the vehicle is long-term invalid and the vehicle is in a complex motion mode, but also has strong anti-vibration performance and is suitable for both electric vehicles with small vibration energy and fuel vehicles with large vibration energy. SUMMARY

[0004] The application aims to provide a vehicle state detection method, medium and equipment to solve the technical problem that the vehicle positioning method in the prior art cannot effectively identify the vehicle motion state in the extreme scene where the GNSS of the vehicle is long-term invalid and the vehicle is in a complex motion mode. The application provides a vehicle state detection method, comprising the following steps: S1, acquiring IMU sensor data installed on a vehicle, the IMU sensor being horizontally placed on the vehicle, and the x, y and z axes of the IMU sensor being installed in the direction satisfying the right-hand rule; The x and z axes are installed in the direction satisfying the right-hand rule; S2, performing low-pass filtering processing on the IMU data, specifically, performing fourth-order Butterworth low-pass filter processing on the IMU data to obtain processed IMU data; S3, multi-scale absolute deviation integral statistical quantity calculation, specifically, calculating the multi-scale absolute deviation integral statistical quantity of long and short windows based on the processed IMU data in S2, and constructing a time series feature vector of the statistical quantity; S4, constructing a TCN time convolution network based on the time series feature vector, training the TCN time convolution network according to label data to obtain a model, and using the trained model to judge the motion state.

[0005] Further improvement of the vehicle state detection method of the application is that when the fourth-order Butterworth low-pass filter processing is performed on the IMU data, the fourth-order Butterworth difference equation is used for processing, and the expression is as follows: ; In the formula, is the result of filtering the first data, denotes the serial number; is the first IMU original data; the coefficient is a feedback term; the coefficient is a weight of the current and historical input to the output.

[0006] Further improvement of the vehicle state detection method of the application is that the calculation of the multi-scale absolute deviation integral statistical quantity of long and short windows based on the processed IMU data in S2 comprises the following steps: S301, calculating the absolute deviation values of long and short windows, and the calculation formula is as follows: ; ; In the formula, is the absolute deviation value of the first long window; the IMU sensor data comprises an accelerometer x-axis data, accelerometer x-axis data, and accelerometer z-axis data; For long window IMU accelerometer The average value of the axis data; For long-window IMU accelerometers The average value of the axis data; To determine the absolute value sign; For the first Absolute deviation value of the short window segment; For the short window IMU accelerometer The average value of the axis data; For short-window IMU accelerometers The average value of the axis data; S302. Add third moments to correct the absolute deviation between the long and short windows. The specific formula is as follows: ; ; In the formula, For the first Absolute deviation of the long window after segment correction; It is the first Standard deviation of segment length window; For the first Segment length window skewness; To represent square; This is the adjustment coefficient; For the first The absolute deviation of the short window after segment correction. It is the first Short window standard deviation For the first The skewness of the short window segment is calculated using the following formula: ; In the formula, For the first Third-order central moment of segment, For the first IMU data, For the first Average value of segment IMU data For sample data size, For the first Segment data skewness For the first Segment IMU data variance; S303. Perform a nonlinear transformation on the deviation value obtained in S302 to obtain the multi-scale absolute deviation integral statistic. The specific transformation formula is as follows: ; ; In the formula, This is the integral statistic of the absolute deviation over a long window and multiple scales. It is a logarithmic function; This is the penalty coefficient; To find the function with the maximum value; This is the integral statistic of absolute deviation across short-window, multi-scale operations.

[0007] S303. Construct a time-series feature vector from the multi-scale absolute deviation integral statistics of long and short windows, specifically arranging the time-series feature vector according to the time sequence. as follows: ; ; in, The first in the time series feature vector 1 eigenvector For the first A long-window, multi-scale absolute deviation integral statistic For the first A short-window, multi-scale absolute deviation integral statistic.

[0008] S304. Using the temporal feature vector obtained from S302, construct a TCN temporal convolutional network model. The network input is the temporal feature vector obtained in S3, and the output of the network model is the still detection label. The network model mainly consists of four residual blocks. Each residual block contains causal convolution, layer normalization, LeakyReLU activation function, and Dropout layer. The causal convolution output depends only on the current and previous inputs, not on future inputs. To increase the receptive field of the model without increasing the parameters, dilated convolution is added on top of the causal convolution. The mathematical model of causal convolution is as follows: ; In the formula, For convolution kernel, The kernel size is [size]. Given a causal convolution input sequence and a temporal feature vector obtained from S3, For serial number, The output of causal convolution, The expansion factor is 2 (here). For the current time step; after the convolution, a normalization layer is added to standardize the output of the causal convolution. The normalization model is as follows: ; In the formula, The mean in the current layer, The variance in the current layer. The output of the causal convolution is used as the input for layer normalization; It is a small constant. This is the layer-normalized output; the activation function used in the network is LeakyReLU, and its model is as follows: ; In the formula, As input, The output is the activation function; to prevent overfitting in the convolutional network, a Dropout layer is added. The model for the Dropout layer is as follows: ; In the formula, Let be a Bernoulli random variable. For the probability of discarding, As input to the Dropout layer Output for Dropout layer; After training the model using labeled data, the model output is used directly to determine the result. A model output of 0 indicates movement, and an output of 1 indicates stillness.

[0009] The present invention also provides a readable storage medium storing a computer program adapted to be loaded by a processor and executed as described above for the vehicle state detection method.

[0010] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, runs the vehicle state detection method as described above.

[0011] The application of the technical solution of the present invention has the following beneficial effects: The vehicle state detection method of this invention preprocesses the original acceleration signal through a fourth-order Butterworth low-pass filter. This filter has the characteristics of maximum flat passband, ensuring that the effective signal passes through the steep roll-off characteristic without distortion. It can effectively suppress high-frequency noise such as engine vibration and road excitation with zero phase distortion design. Through bidirectional filtering, it avoids signal time shift, thereby eliminating high-frequency vibration noise of the vehicle, reducing the impact of noise on stationary detection, and realizes vehicle stationary detection by utilizing multi-scale absolute deviation consistency. The design of the multi-scale absolute deviation consistency index is as follows: ① Utilizing the multi-segment consistency of the data can effectively identify the stationarity of time series data and is unaffected by the magnitude of vehicle vibration, exhibiting strong vibration resistance; ② Incorporating variance and skewness can effectively capture the impact of local asymmetry of the data on the overall index, helping to identify the state of the vehicle in extreme scenarios with complex motion patterns; ③ Incorporating a maximum deviation penalty term highlights the impact of the maximum deviation, assisting in identifying the vehicle in a continuous low-speed linear motion field; ④ Using long and short window statistics to construct statistical time-series features, long window statistics are used for continuous and stable stationary states, while short window statistics are used to capture instantaneous and brief stationary states, thus compensating for the problem of long windows being unable to respond in a timely manner due to the sliding window duration; ⑤ Utilizing the TCN temporal convolutional network model to learn the motion state patterns from the long and short window statistics, thereby detecting stationary states.

[0012] This invention can effectively identify the vehicle motion state under continuous low-speed straight-line or frequent start-up scenarios when GNSS is lost for a long time, solving the technical problem that existing vehicle positioning methods cannot effectively identify the vehicle motion state in extreme scenarios where the vehicle's GNSS is lost for a long time and the vehicle is in a complex motion mode; moreover, this invention has strong vibration resistance, is not affected by the magnitude of vehicle vibration energy, and can be used in both electric vehicles and gasoline vehicles.

[0013] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the vehicle state detection method of the present invention; Figure 2 This is the first set of data, showing the results of the traditional static detection method. Figure 3 This is a graph showing the detection results of the first set of data using the vehicle condition detection method of this invention; Figure 4This is the result image of the second set of data detected using the traditional static detection method; Figure 5 This is a graph showing the detection results of the second set of data using the vehicle condition detection method of this invention. Detailed Implementation

[0015] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0016] See Figure 1 As shown, a vehicle condition detection method includes the following steps: S1. Acquire data from the IMU (Inertial Measurement Unit) sensor installed on the vehicle; the IMU sensor data is raw data at 100Hz; the IMU sensor is placed horizontally on the vehicle, and the IMU sensor's... The axes are oriented in the direction of vehicle movement, and the x and z axes are installed in accordance with the right-hand rule. S2. Perform low-pass filtering on the IMU data. Specifically, perform a fourth-order Butterworth low-pass filter on the IMU data to obtain processed IMU data with a cutoff frequency of 40Hz and a sampling rate of 100Hz. This eliminates high-frequency vibration noise from the vehicle and reduces the impact of noise on stationary detection. Specifically, when processing IMU data using a fourth-order Butterworth low-pass filter, the process is performed using a fourth-order Butterworth difference equation, as shown in the following expression: ; In the formula, For the first The result after data filtering Indicates the serial number; For the first Raw data from IMUs; coefficients For feedback items; coefficients To control the weights of current and historical inputs on the output.

[0017] S3. Multi-scale absolute deviation integral statistics for stationary detection: Specifically, based on the IMU data processed in S2, multi-scale absolute deviation integral statistics with long and short windows are calculated, and a time-series feature vector of the statistics is constructed. This highlights the influence of deviation and facilitates further differentiation between stationary and moving states. Long and short window stationary detection conditions are used to determine the vehicle's state, reducing missed detections caused by the long window. In this embodiment, the long window is 5 seconds, and the short window is 2 seconds.

[0018] Preferably, when calculating the multi-scale absolute deviation integral statistic based on the IMU data processed in S2, the following steps are included: S301. The long window uses 5 seconds of data, divided into five segments of 1 second each for calculation. The short window uses 2 seconds of data, divided into four segments of 0.5 seconds each for calculation. Calculate the absolute deviation between the long and short windows using the following formula: ; ; In the formula, For the first Segment length window absolute deviation value; IMU sensor data including accelerometer x-axis data, accelerometer x-axis data, and accelerometer z-axis data; For long window IMU accelerometer The average value of the axis data; For long window 5-second IMU accelerometer The average value of the axis data; To determine the absolute value sign; For the first Absolute deviation value of the short window segment; For the short window IMU accelerometer The average value of the axis data; For short-window 2-second IMU accelerometer The average value of the axis data.

[0019] S302. Add third moments to correct the absolute deviation between the long and short windows. The specific formula is as follows: ; ; In the formula, For the first Absolute deviation of the long window after segment correction; It is the first Standard deviation of segment length window; For the first Segment length window skewness; To represent square; The adjustment coefficient is set to 0.5. For the first The absolute deviation of the short window after segment correction. It is the first Short window standard deviation For the first The skewness of the short window segment is calculated using the following formula: ; In the formula, For the first Third-order central moment of segment, For the first IMU data, For the first Average value of segment IMU data For sample data size, For the first Segment data skewness For the first Variance of segment IMU data.

[0020] S303. Perform a nonlinear transformation on the deviation value obtained in S302 to obtain the multi-scale absolute deviation integral statistic. The specific transformation formula is as follows: ; ; In the formula, This is the integral statistic of the absolute deviation over a long window and multiple scales. It is a logarithmic function; The penalty coefficient is set to 0.2. To find the function with the maximum value; This is the integral statistic of absolute deviation across short-window, multi-scale operations.

[0021] S304. Construct a time-series feature vector from the multi-scale absolute deviation integral statistics of long and short windows, specifically arranging the time-series feature vector according to the time sequence. as follows: ; ; in, The first in the time series feature vector 1 eigenvector For the first A long-window, multi-scale absolute deviation integral statistic For the first A short-window, multi-scale absolute deviation integral statistic.

[0022] S4 constructs a TCN temporal convolutional network based on temporal feature vectors, trains the TCN temporal convolutional network according to the label data to obtain the model, and uses the trained model to determine the motion state. Specifically, a TCN temporal convolutional network model is first constructed. The input to the network is the temporal feature vector obtained in the previous step, and the output of the network model is the still detection label. The network model mainly consists of four residual blocks. Each residual block contains causal convolution, layer normalization, LeakyReLU activation function (linear rectified function), and Dropout layer. The output of the causal convolution only depends on the current and previous inputs, not on future inputs. In order to increase the receptive field of the model without increasing the number of parameters, a dilated convolution is added on the causal convolution. The mathematical model of causal convolution is as follows: ; In the formula, For convolution kernel, The kernel size is [size]. The input sequence for the causal convolution is given by [the input sequence], and the feature vector obtained in the previous step is given by [the input sequence]. For serial number, The output of causal convolution, The expansion factor is 2 (here). For the current time step; after the convolution, a normalization layer is added to standardize the output of the causal convolution. The normalization model is as follows: ; In the formula, This is the mean value in the current layer. The variance in the current layer. The output of the causal convolution is used as the input for layer normalization. It is a small constant. This is the layer-normalized output; the activation function used in the network is LeakyReLU, and its model is as follows: ; In the formula, As input, The output is the activation function; to prevent overfitting in the convolutional network, a Dropout layer is added. The model for the Dropout layer is as follows: ; In the formula, For Bernoulli random variables (in terms of probability) Take 0, 1− Take 1). For the probability of discarding, As input to the Dropout layer This is the output of the Dropout layer, and is usually set to 0.5. After training the model with a large amount of labeled data, the model output is used directly to determine the result. The model output of 0 indicates movement, and the output of 1 indicates stillness.

[0023] The present invention also provides a readable storage medium storing a computer program adapted to be loaded by a processor and executed as described above for the vehicle state detection method.

[0024] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, runs the vehicle state detection method as described above.

[0025] To verify the performance of the stationary detection algorithm proposed in this invention, two sets of vehicle-mounted data from different powertrain systems were collected in the same scenario. The IMU sensor sampling frequency was 100Hz, and a high-precision fiber optic inertial navigation system was used as the ground truth. The vehicle was slowly traveling in a straight line in a GNSS-free environment to test the stationary detection performance in a low-speed straight-line scenario without GNSS. The first and second sets of data were collected from electric vehicles and gasoline vehicles, respectively, and the stationary detection performance on electric vehicles with lower vibration energy and gasoline vehicles with higher vibration energy was compared. Figures 2-5 The detection results of two stationary detection methods on two datasets are given respectively, where "0" indicates that the vehicle is in motion and "1" indicates that the vehicle is stationary.

[0026] To verify the static detection performance of this invention, false detection rate, false negative rate, and accuracy rate were used as evaluation indicators, and the positioning of each indicator is as follows: ; ; ; In the formula, False positive rate; The false negative rate; For accuracy; This represents the total epoch time during which the system was actually at rest. This represents the total epoch time of the actual state of motion. This represents the total epoch time during which a stationary position was detected, but the actual movement occurred. This represents the total epoch time during which the detection is dynamic but actually static. This represents the total epoch time during which the detected state was static when it was actually static. This indicates the total epoch time during which the detection is dynamic but actually dynamic.

[0027] As can be seen from the first set of data in Table 1, the false detection rate of the method of this invention is significantly lower than that of the traditional method. The static detection accuracy of this invention is 99.43%, while that of the traditional method is 74.51%, representing an improvement of 34.45%. This indicates that the static detection accuracy of the method of this invention is significantly improved in scenarios with low vibration, such as trams and low-speed straight lines without GNSS. As can be seen from the second set of data in Table 2, the false detection rate of the method of this invention is significantly lower than that of the traditional method. The static detection accuracy of this invention is 99.27%, while that of the traditional method is 66.71%, representing an improvement of 48.81%. This indicates that the static detection accuracy of the method of this invention is significantly improved in scenarios with high engine vibration, such as gasoline vehicles and low-speed straight lines without GNSS.

[0028] Table 1. Results of static detection using two methods for the first set of data.

[0029] Table 2. Results of static detection using two methods for the second set of data.

[0030] This invention relates to a vehicle state detection method that preprocesses the original acceleration signal using a fourth-order Butterworth low-pass filter. This filter has a maximum flat passband characteristic, ensuring that the effective signal passes through a steep roll-off characteristic without distortion. It can effectively suppress high-frequency noise such as engine vibration and road excitation with zero phase distortion. Through bidirectional filtering, it avoids signal time shift, thereby eliminating high-frequency vibration noise of the vehicle and reducing the impact of noise on stationary detection. Furthermore, it utilizes multi-scale absolute deviation consistency to achieve vehicle stationary detection. This method can effectively identify the vehicle motion state in scenarios of continuous low-speed straight-line movement or frequent starts under long-term GNSS lockout, solving the technical problem that existing vehicle positioning methods cannot effectively identify the vehicle motion state in extreme scenarios where the vehicle's GNSS is unavailable for a long time and the vehicle is in a complex motion mode. Moreover, this invention has strong vibration resistance, is not affected by the magnitude of vehicle vibration energy, and can be used in both electric and gasoline vehicles.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A vehicle state detection method characterized by comprising: Comprise the following steps: S1, acquiring IMU sensor data installed on the vehicle, the IMU sensor is horizontally placed on the vehicle, and the x and z axes of the IMU sensor are installed towards the right-hand rule axis towards the direction of vehicle movement, the x and z axes towards the right-hand rule S2, low-pass filter processing is carried out on the IMU data, specifically: the IMU data is processed by a fourth-order Butterworth low-pass filter to obtain processed IMU data; S3, multi-scale absolute deviation integral statistics calculation, specifically: based on the processed IMU data in S2, the multi-scale absolute deviation integral statistics of long and short windows are calculated, and a time series feature vector is constructed; S4, a TCN time convolution network is constructed based on the time series feature vector, and a model is obtained by training the TCN time convolution network according to label data, and the trained model is used to judge the motion state.

2. The vehicle state detection method according to claim 1, characterized by, When the IMU data is processed by a fourth-order Butterworth low-pass filter, it is processed by a fourth-order Butterworth difference equation, and the expression is as follows: ； In the formula, is the result of the data filtering, is the result of the data filtering, represents the serial number; is the result of the data filtering, is the result of the data filtering, is the feedback term; coefficient is the control of the current and historical input on the output weight.

3. The vehicle state detection method according to claim 1, characterized by, Based on the processed IMU data in S2, the long and short window multi-scale absolute deviation integral statistics are calculated, comprising the following steps: S301, calculate the absolute deviation value of the long window and the short window, and the calculation formula is as follows: ； ； wherein is the first segment absolute deviation value; the IMU sensor data includes accelerometer axis data, accelerometer x-axis data, and accelerometer z-axis data; is the average of the IMU accelerometer axis data for the first segment of the long window; is the average of the IMU accelerometer axis data for the long window; is the absolute value function; is the first segment absolute deviation value for the short window; is the average of the IMU accelerometer axis data for the first segment of the short window; is the average of the IMU accelerometer axis data for the short window; S302, add the third moment to correct the absolute deviation of the long window and the short window, and the specific formula is as follows: ； ； In the formula, For the first Absolute deviation of the long window after segment correction; It is the first Standard deviation of segment length window; For the first Segment length window skewness; To represent square; This is the adjustment coefficient; For the first The absolute deviation of the short window after segment correction. It is the first Short window standard deviation For the first The skewness of the short window segment is calculated using the following formula: ； wherein, is the first segment third central moment, is the first segment IMU data, is the first segment IMU data mean, is the sample data quantity, is the first segment data skewness, is the first segment IMU data variance; S303, the deviation value obtained by S302 is subjected to nonlinear transformation to obtain multi-scale absolute deviation integral statistics, and the specific change formula is as follows: ； ； wherein is a long window multiscale absolute deviation integral statistic; is a logarithm function; is a penalty coefficient; is a maximum function; is a short window multiscale absolute deviation integral statistic.

4. The vehicle state detection method according to claim 1, characterized by S3 further comprises S304, constructing a time sequence feature vector for the long-short window multi-scale absolute deviation integral statistics, specifically, a time sequence feature vector is formed according to time sequence, and a specific time sequence feature vector is As follows: ； ； in, The first in the time series feature vector 1 eigenvector For the first A long-window, multi-scale absolute deviation integral statistic For the first A short-window, multi-scale absolute deviation integral statistic.

5. The vehicle state detection method according to claim 4, characterized by The TCN time convolution network model is constructed, the input of the network is the time series feature vector obtained by S3, the output of the network model is the static detection label, and the network model mainly consists of four residual blocks. Each residual block contains causal convolution, layer normalization, LeakyReLU activation function and Dropout layer. The mathematical model of causal convolution is as follows: ； In the formula, is a convolution kernel, is a convolution kernel size, is a causal convolution input sequence, that is, a time sequence feature vector obtained by S3, is a serial number, is an output of the causal convolution, is an expansion factor, is a current time; after convolution, layer normalization is added, and the output of the causal convolution is standardized. The model of the layer normalization is as follows: ； where, the mean in the current layer, the variance in the current layer, is the output of the causal convolution, which serves as the layer normalization input; is a small constant, is the layer normalization output; the activation function used in the network is LeakyReLU, which is modeled as follows: ； In the formula, As input, is the activation function output; in order to prevent the convolutional network from overfitting, a Dropout layer is added, and the model of the Dropout layer is as follows: ； wherein is a Bernoulli random variable, is a dropout probability, as input to the Dropout layer, is the output of the Dropout layer; After the model is trained using label data, the output of the model is directly used to judge the result. The model output is 0, indicating motion, and the output is 1, indicating static.

6. A readable storage medium characterized by, The readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by the processor to execute the vehicle state detection method of any one of claims 1-5.

7. A computer device, characterized by The computer device comprises a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to run the vehicle state detection method of any one of claims 1-5.

Images