Vehicle collision identification method and device

Abstract

This application provides a vehicle collision recognition method and apparatus, comprising: obtaining an acceleration signal and a piezoelectric signal from a vehicle; determining the vehicle's travel speed when the acceleration signal and the piezoelectric signal are obtained; determining a target signal from a first collision signal and a second collision signal based on the travel speed; correlating the first collision signal with the acceleration signal; and correlating the second collision signal with the piezoelectric signal; and determining a vehicle collision recognition result based on the target signal. This improves the accuracy of the collision recognition result.

Description

Technical Field

[0001] The present application relates to the field of vehicle safety, and in particular to a vehicle collision identification method and device. Background Art

[0002] Vehicle micro-collision recognition technology is a key technology in the automotive safety field. It can be used to monitor the status of parked vehicles and also to detect collisions while driving, especially at low speeds. Related technologies use real-time camera image capture to identify vehicle collisions. Artificial intelligence (AI) technology identifies approaching objects or image jitter as a vehicle being impacted. However, this solution suffers from low accuracy in identifying vehicle collisions, making it prone to misjudgment. Summary of the Invention

[0003] The present application mainly provides a vehicle collision recognition method and device, which can overcome the problem of low accuracy in identifying vehicle collisions in related technologies.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] The present invention provides a method for identifying a vehicle collision, including:

[0006] Obtaining the vehicle's acceleration signal and piezoelectric signal;

[0007] determining a travel speed of the vehicle when acquiring the acceleration signal and the piezoelectric signal;

[0008] determining a target signal in the first collision signal and the second collision signal based on the driving speed;

[0009] A collision recognition result of the vehicle is determined based on the target signal.

[0010] An embodiment of the present application provides a vehicle collision recognition device, the device comprising:

[0011] An acquisition unit, used to acquire an acceleration signal and a piezoelectric signal of the vehicle;

[0012] a first determining unit, configured to determine a travel speed of the vehicle when acquiring the acceleration signal and the piezoelectric signal;

[0013] a second determining unit, configured to determine a target signal from the first collision signal and the second collision signal based on the driving speed;

[0014] The recognition unit is configured to determine a collision recognition result of the vehicle based on the target signal.

[0015] An embodiment of the present application provides a vehicle collision recognition device, comprising: a memory for storing executable instructions; and a processor for implementing the vehicle collision recognition method provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0016] An embodiment of the present application provides a storage medium having executable instructions stored thereon. When the executable instructions are executed by a processor, the method for vehicle collision identification provided in the embodiment of the present application is implemented.

[0017] The embodiments of the present application have the following beneficial effects:

[0018] The embodiment of the present application obtains the acceleration signal and piezoelectric signal of the vehicle; determines the vehicle's driving speed when the acceleration signal and piezoelectric signal are obtained; based on the driving speed, selects a target signal from a first collision signal related to the acceleration signal and a second collision signal related to the piezoelectric signal; and determines the vehicle's collision recognition result based on the target signal. In this way, the present application can dynamically select a collision signal that matches the vehicle's current driving speed, and finally determine the vehicle's collision recognition result based on the selected collision signal. In this way, by determining the vehicle's collision recognition result using a collision signal that matches the driving speed, the accuracy of identifying vehicle collisions can be improved, thereby avoiding misjudgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a vehicle collision recognition method according to an embodiment of the present invention;

[0020] Figure 2 This embodiment of the present application provides Figure 1 Schematic diagram of the process of S104;

[0021] Figure 3 This embodiment of the present application provides Figure 2 Schematic diagram of the process of S202;

[0022] Figure 4 This embodiment of the present application provides Figure 2 Schematic diagram of the process of S204;

[0023] Figure 5 A schematic diagram of a vehicle collision recognition method according to an embodiment of the present invention;

[0024] Figure 6 A schematic diagram of the implementation framework of the vehicle collision recognition method provided in an embodiment of the present application;

[0025] Figure 7 A scenario framework diagram of the vehicle collision recognition method provided in an embodiment of the present application;

[0026] Figure 8 A schematic diagram of the structure of a vehicle collision recognition device provided in an embodiment of the present application;

[0027] Figure 9 A schematic diagram of the structure of a vehicle collision recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solution of the present application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] In order to enable people in this technical field to better understand the embodiments of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of this application, rather than all the embodiments.

[0030] The terms "first," "second," and "third," etc. in the specification, embodiments, claims, and figures of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, for example, including a series of steps or units. A method, system, product, or apparatus is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to the process, method, product, or apparatus.

[0031] Currently, there are four technologies used for status monitoring of parked vehicles: 1. Real-time acquisition of camera images, and use of AI technology to identify objects approaching or when the image shakes, to determine that the vehicle has been hit. The disadvantage of this method is that the camera and its corresponding controller need to be supplied with constant power, which consumes a lot of power. At the same time, when the object attacking the vehicle is indistinguishable by AI technology and the damage is minor (the vehicle has no vibration), it cannot be identified; 2. Use acceleration sensors to identify collisions, but the resolution of acceleration sensors is low, and the accuracy of micro-collision recognition is low. It can only be accurately identified when the collision is more serious; 3. Use pressure sensors to collect the pressure generated by vehicle collisions to identify collisions, but this technology can only Identify areas close to sensors, which requires sensors to be placed all over the car body, which is costly and difficult to engineer. 4. Identify micro-collisions by using piezoelectric sensors to identify the piezoelectric signals generated by elastic waves caused by micro-collisions. This method can identify very weak collisions, and can even identify collisions as small as "tapping the car body with your hand." However, the disadvantage of this method is that it is too sensitive and has a high false alarm rate. For example, rain, snow, leaves falling on the car body, vibrations generated by heavy vehicles passing by, and horns will all be identified. The current recognition algorithm cannot accurately distinguish between micro-collision scenarios that need to be recorded and reported to the owner and erroneous scenarios that do not need to be reported to the owner.

[0032] In order to solve the above technical problems, an embodiment of the present application provides a vehicle collision recognition method, which can be applied to a vehicle collision recognition device of a vehicle.

[0033] Figure 1 This is a flow chart of the vehicle collision recognition method according to an embodiment of the present application. Figure 1 As shown, the process may include:

[0034] In S101 , an acceleration signal and a piezoelectric signal of the vehicle are acquired.

[0035] Here, the acceleration signal refers to a time domain signal of the collision acceleration of the vehicle that varies with time; the piezoelectric signal refers to a time domain signal of the piezoelectricity generated when the vehicle vibrates that varies with time.

[0036] In some embodiments, the acceleration signal and the piezoelectric signal can be collected by an acceleration sensor module and a piezoelectric sensor module respectively deployed on the vehicle, and then the acceleration sensor module and the piezoelectric sensor module send the collected collision signals to the vehicle collision recognition device.

[0037] In S102 , the driving speed of the vehicle when the acceleration signal and the piezoelectric signal are acquired is determined.

[0038] Here, driving speed refers to the vehicle's current driving speed when the acceleration sensor module and the piezoelectric sensor module respectively acquire acceleration and piezoelectric signals. Specifically, driving speed refers to the vehicle's driving speed when a collision occurs, causing the sensor module to acquire the collision signal. In other words, the time for determining the vehicle's driving speed is the same as or approximately the same as the time for acquiring the acceleration and piezoelectric signals.

[0039] During actual application, when collecting collision signals, the sensor module will send the collected collision signals and the timestamp corresponding to the collision signals to the vehicle collision recognition device. After receiving the timestamp corresponding to the collision signal, the vehicle collision recognition device determines the timestamp at which the acceleration signal and the piezoelectric signal are the same, and then obtains the vehicle's driving speed at the timestamp.

[0040] In S103 , a target signal is determined between the first collision signal and the second collision signal based on the driving speed.

[0041] Here, the first collision signal is related to the acceleration signal, and the second collision signal is related to the piezoelectric signal; wherein, the first collision signal is related to the acceleration signal, and the second collision signal is related to the piezoelectric signal means that the vehicle collision recognition device can use the first collision signal as the acceleration signal and the second collision signal as the piezoelectric signal, that is, the vehicle collision recognition device can directly select the target signal for determining the vehicle collision recognition result from the acceleration signal and the piezoelectric signal based on the driving speed; in some embodiments, the first collision signal is related to the acceleration signal may also mean that the acceleration signal is calculated to obtain the first collision signal, and the second collision signal is related to the piezoelectric signal means that the piezoelectric signal is calculated to obtain the second collision signal.

[0042] In some embodiments, S103 may include: when the driving speed is greater than a preset driving speed, using a first collision signal calculated from the acceleration signal as a target signal or determining the acceleration signal as a target signal; when the driving speed is less than or equal to the preset driving speed, using a second collision signal calculated from the piezoelectric signal as a target signal or determining the piezoelectric signal as a target signal.

[0043] Here, the preset driving speed may be a driving speed value automatically set by the vehicle collision recognition device for determining the vehicle state. The preset driving speed may be used to determine whether the vehicle is in a dynamic driving state or a quasi-static state.

[0044] In some embodiments, when the target signal is obtained based on calculation of an acceleration signal or a piezoelectric signal, S103 may include: determining a changing trend of an acceleration characteristic of a partial signal of the acceleration signal, and taking the changing trend of the acceleration characteristic as a first collision signal; determining a total energy value of the piezoelectric characteristic in a partial signal of the piezoelectric signal; based on the total energy value, determining a changing trend of the piezoelectric characteristic in a partial signal of the piezoelectric signal, and taking the changing trend of the piezoelectric characteristic as a second collision signal; when the driving speed is greater than a preset driving speed, determining the first collision signal as a target signal; when the driving speed is less than or equal to the preset driving speed, determining the second collision signal as a target signal.

[0045] In some embodiments, when the target signal is obtained by selecting from an acceleration signal and a piezoelectric signal (i.e., the first collision signal is an acceleration signal and the second collision signal is a piezoelectric signal), the vehicle collision recognition device can determine whether the vehicle is in dynamic driving based on the driving speed. If the vehicle is determined to be in dynamic driving based on the driving speed, the acceleration signal can be determined as the target signal to determine the vehicle collision recognition result through the acceleration signal; if the vehicle is determined to be not in dynamic driving based on the driving speed (i.e., the vehicle is in a quasi-static environment), the piezoelectric signal can be determined as the target signal to determine the vehicle collision recognition result through the piezoelectric signal. This is because when a vehicle in dynamic driving collides, the acceleration generated by the collision is large, so the vehicle collision recognition result is determined by the piezoelectric signal; when a vehicle in a quasi-static environment collides, the acceleration generated by the collision is small. If the vehicle collision recognition result is determined by the acceleration signal, the accuracy of identifying the vehicle collision is reduced. Therefore, the vehicle collision recognition result is determined by the piezoelectric signal.

[0046] After acquiring the acceleration signal, piezoelectric signal, and vehicle speed, the embodiment of the present application can determine a target signal for determining a vehicle collision identification result based on the acceleration signal when the vehicle speed is greater than a preset speed; and determine the target signal based on the piezoelectric signal when the vehicle speed is less than or equal to the preset speed. In this way, the vehicle collision identification result can be determined based on the acceleration signal when the vehicle is dynamically traveling, and based on the piezoelectric signal when the vehicle is quasi-static, thereby improving the accuracy of vehicle collision identification.

[0047] In S104 , a collision recognition result of the vehicle is determined based on the target signal.

[0048] Here, the collision recognition result is used to indicate whether the vehicle has collided and the degree of collision if a collision has occurred.

[0049] In some embodiments, the vehicle collision recognition device can input a target signal into a trained first neural network model, and the trained first neural network model outputs a vehicle collision recognition result, where the target signal is a piezoelectric signal or an acceleration signal.

[0050] In some embodiments, when the target signal is selected from the acceleration signal and the piezoelectric signal, in order to improve the efficiency and accuracy of determining the collision recognition result, the vehicle collision recognition device can first obtain a partial signal from the target signal, and determine the probability of vehicle collision based on the partial signal. If the probability of vehicle collision indicates that the possibility of vehicle collision is large, then the vehicle collision recognition result is determined based on the target signal; if the probability of vehicle collision indicates that the possibility of vehicle collision is small, there is no need to determine the vehicle collision recognition result.

[0051] In some embodiments, when the target signal is obtained based on calculation of an acceleration signal or a piezoelectric signal, S104 may include: when the target signal is a changing trend of an acceleration characteristic, and the changing trend of the acceleration characteristic is greater than a preset changing trend, determining the collision recognition result of the vehicle based on the acceleration signal; when the target signal is a changing trend of a piezoelectric characteristic, and the changing trend of the piezoelectric characteristic is greater than a preset changing trend, determining the collision recognition result of the vehicle based on the piezoelectric signal.

[0052] In the embodiment of the present application, the acceleration signal and piezoelectric signal of the vehicle are obtained; the driving speed of the vehicle when the acceleration signal and the piezoelectric signal are obtained is determined; based on the driving speed, a target signal is determined in a first collision signal related to the acceleration signal and a second collision signal related to the piezoelectric signal; based on the target signal, the collision recognition result of the vehicle is determined. In this way, the present application can obtain different collision signals, and then dynamically determine the collision signal that matches the driving speed based on the current driving speed of the vehicle, and finally determine the collision recognition result of the vehicle based on the selected collision signal. In this way, by determining the collision recognition result of the vehicle by the collision signal that matches the driving speed, the accuracy of identifying vehicle collisions can be improved, thereby avoiding misjudgments.

[0053] Figure 2 This embodiment of the present application provides Figure 1 The flow diagram of S104, Figure 1 S104 in can be realized by S201 to S204, which will be combined Figure 2 The steps shown are explained.

[0054] In S201 , based on a first portion of the target signal, a change trend of a collision feature of the first portion of the target signal is determined.

[0055] Here, the target signal is an acceleration signal or a piezoelectric signal; the change trend is used to characterize the change trend of the collision feature within the acquisition time range of the first part of the signal. The change trend of the collision feature can be used to more accurately determine the probability of vehicle collision.

[0056] In an embodiment of the present application, the vehicle collision recognition device may first collect a first portion of signals including multiple continuous window signals on a target signal based on a preset signal window, and then determine a change trend of the collision characteristics of the first portion of signals.

[0057] Here, the window width and sliding window step of the preset signal window are preset values, which can be automatically set by the vehicle collision recognition device or pre-set by the user in the vehicle collision recognition device.

[0058] In some embodiments, the vehicle collision recognition device may first determine a sub-variation trend of the collision characteristic of each window signal in the first portion of the signal, and then determine a variation trend of the collision characteristic of the first portion of the signal based on the sub-variation trend of the collision characteristic of each window signal. The collision characteristic may be an acceleration characteristic in the acceleration signal or a piezoelectric characteristic in the piezoelectric signal.

[0059] In some embodiments, when the target signal is a piezoelectric signal, the vehicle collision recognition device may first determine the total energy value of the piezoelectric feature in the first part of the signal; and then determine the change trend of the piezoelectric feature in the first part of the signal based on the total energy value.

[0060] In S202 , the collision probability of the vehicle is determined based on the change trend of the collision characteristics of the first portion of signals.

[0061] Here, a vehicle's collision probability represents the likelihood of a collision. When the collision probability meets the pre-set collision conditions, the likelihood of a collision is high, and further assessment of the vehicle's collision severity is required. When the collision probability does not meet the pre-set collision conditions, the likelihood of a collision is low, and no further assessment is required.

[0062] In an embodiment of the present application, the vehicle collision recognition device can calculate the changing trend of the collision characteristics of the first signal portion and determine the collision probability of the vehicle based on the changing trend of the collision characteristics. Because the changing trend of the collision characteristics can represent the changing state of the collision characteristics, when the changing trend is large, it indicates that the collision characteristics of the first signal portion are changing in a large direction, and the probability of the vehicle collision is high.

[0063] In S203 , when the collision probability satisfies a preset collision condition, a second portion of the signal is obtained from the target signal.

[0064] In the embodiment of the present application, when the vehicle collision recognition device determines that the collision probability meets the preset collision condition, it means that the possibility of the vehicle collision is high, so it is necessary to obtain the second part of the signal in the target signal to further determine the vehicle collision recognition result.

[0065] In some embodiments, "obtaining the second portion of the signal from the target signal" in S203 can be implemented through S231 and S232: In S231, the difference between the current time information and the preset time period is used as the start acquisition time, and the current time information is used as the end acquisition time. In S232, based on the start acquisition time and the end acquisition time, the second portion of the signal is obtained from the target signal.

[0066] In an embodiment of the present application, the vehicle collision recognition device can first obtain current time information, then use the difference between the current time information and a preset time period as the start time for collection, and the current time information as the end time for collection. This allows the device to determine the time period for collecting the second portion of the signal. In actual applications, when the vehicle collision recognition device determines that the collision probability meets the preset collision conditions, it indicates that the vehicle may have collided, and therefore it is necessary to collect the collision signal before the collision occurs, i.e., the second portion of the signal. This can improve the accuracy of the collision recognition result.

[0067] In S204 , a collision recognition result of the vehicle is determined based on the second partial signal.

[0068] In an embodiment of the present application, after obtaining the second partial signal from the target signal, the vehicle collision recognition device may process the second partial signal and determine a vehicle collision recognition result based on the processed second partial signal.

[0069] Here, the processing of the second portion of the signal may be a filtering and transforming process. In some embodiments, the second portion of the signal after filtering and transforming may be further processed, where the processing method may be a range conversion process. The range conversion process can obtain signals of different dimensions, thereby improving the accuracy of the vehicle collision recognition result.

[0070] In some embodiments, the vehicle collision recognition device may process the second portion of the signal, then perform feature extraction on the processed second portion of the signal to obtain typical features that can reflect the signal waveform, and finally determine the vehicle collision recognition result based on the extracted features.

[0071] In an embodiment of the present application, the vehicle collision recognition device first obtains a first portion of the target signal and determines the vehicle's collision probability based on the first portion of the target signal. If the collision probability satisfies a preset collision condition, the device obtains a second portion of the target signal and determines the vehicle's collision recognition result based on the second portion of the target signal. Thus, in an embodiment of the present application, the device determines the vehicle's collision recognition result based on the second portion of the target signal only after determining that the collision probability satisfies the preset collision condition. This improves the accuracy of the vehicle's collision recognition result.

[0072] Figure 3This embodiment of the present application provides Figure 2 The flow chart of S201 is shown in the figure. Figure 2 S201 in the above can be realized by S301 to S303, which will be combined with Figure 3 The steps shown are explained.

[0073] In S301 , for each window signal, a sub-change trend of a collision feature of the window signal is determined.

[0074] In an embodiment of the present application, the vehicle collision recognition device can collect a first part of the signal including at least two consecutive window signals on the target signal based on a preset signal window, and then determine the sub-change trend of the collision feature of the window signal for each window signal.

[0075] Here, the sub-change trend is used to characterize the change trend of the collision feature within the acquisition time range of the window signal.

[0076] In some embodiments, when the target signal is an acceleration signal, the collision feature in the acceleration signal is an acceleration feature, and S301 can be implemented by the following steps:

[0077] In S311 , for each window signal, at least two acceleration features are obtained on the window signal.

[0078] In an embodiment of the present application, an acceleration time curve can be synthesized based on the acceleration signal, and then the acceleration time curve can be generated. The acceleration sensor can collect the acceleration in the X, Y, and Z directions corresponding to each time point, and based on the acceleration in the X, Y, and Z directions corresponding to each time point, the acceleration characteristics corresponding to each time point on the acceleration time curve can be determined, and then the acceleration time curve can be generated based on the acceleration characteristics corresponding to each time point. In some embodiments, the acceleration characteristics can be implemented by formula (1):

[0079]

[0080] Among them, a i is the acceleration feature in the i-th window signal, a ix 、a iy 、a iz are the acceleration values ​​in the X, Y, and Z directions collected by the acceleration sensor in the i-th window signal.

[0081] In S312 : based on at least two acceleration features and a time interval between two adjacent acceleration features, a sub-change trend of the acceleration feature of the window signal is determined.

[0082] In an embodiment of the present application, the vehicle collision recognition device can determine the slope of each of the two adjacent acceleration features on the window signal based on at least two acceleration features and the time interval between the two adjacent acceleration features, and then calculate the average of the slopes of each of the two adjacent acceleration features on the window signal to obtain a sub-change trend of the acceleration feature of the window signal. In some embodiments, the sub-change trend can be implemented using formula (2):

[0083]

[0084] Among them, k i is the sub-change trend of the i-th window signal, a i is the acceleration feature in the i-th window signal, T is a i+1 with a i The time interval between.

[0085] In some embodiments, when the target signal is a piezoelectric signal, the collision feature in the piezoelectric signal is a piezoelectric feature. In this case, S301 can also be implemented by the following steps:

[0086] In S313 , for each window signal, a total energy value of the piezoelectric feature in the window signal is determined.

[0087] In S314 , for each window signal, based on the total energy value, the energy mean of the collision feature in the window signal is determined to obtain a sub-variation trend of the piezoelectric feature of the window signal.

[0088] In an embodiment of the present application, a vehicle collision recognition device may first generate a piezoelectric curve based on a time-varying piezoelectric signal acquired by a piezoelectric sensor to obtain a piezoelectric curve equation. The piezoelectric curve equation is then processed to calculate the area of ​​the curve corresponding to the processed piezoelectric curve equation over the time region of the acquired window signal to obtain the total energy value of the piezoelectric characteristic in the window signal. The device then calculates the average area of ​​the curve over the time region of the acquired window signal to obtain the mean energy of the collision characteristic in the window signal, thereby obtaining a sub-trend of the piezoelectric characteristic of the window signal. In some embodiments, the piezoelectric curve equation may be processed by squaring the piezoelectric curve equation.

[0089] In S302 , statistical processing is performed on at least two sub-change trends to obtain a change trend of the collision characteristics of the first part of the signal.

[0090] In the embodiment of the present application, the vehicle collision recognition device averages at least two sub-change trends to obtain the change trend of the collision characteristics of the first part of the signal.

[0091] In some embodiments, when the first partial signal is a partial signal of the acceleration signal, the variation trend of the acceleration characteristic of the first partial signal can be realized by formula (3):

[0092]

[0093] Among them, k aver refers to the changing trend of acceleration characteristics; m refers to the number of window signals; k i It refers to the sub-changing trend of the acceleration feature of the i-th window signal.

[0094] In some embodiments, when the first partial signal is a partial signal of a piezoelectric signal, the variation trend of the piezoelectric characteristic of the second partial signal can be realized by formula (4):

[0095]

[0096] Among them, E aver refers to the changing trend of the piezoelectric characteristics; m refers to the number of window signals; E i It refers to the sub-change trend of the piezoelectric characteristics of the i-th window signal.

[0097] In some embodiments, when the target signal is a piezoelectric signal, the collision feature in the piezoelectric signal is a piezoelectric feature. In this case, the above S201 can also be implemented through S221 to S222:

[0098] In S221 , the total energy value of the piezoelectric feature in the first portion of the signal is determined.

[0099] In S222: based on the total energy value, determine the change trend of the piezoelectric characteristic in the first part of the signal.

[0100] In an embodiment of the present application, a vehicle collision recognition device may first generate a piezoelectric curve based on a time-varying piezoelectric signal acquired by a piezoelectric sensor to obtain a piezoelectric curve equation. The piezoelectric curve equation is then processed to calculate the area of ​​the curve corresponding to the processed piezoelectric curve equation over the time period during which the first portion of the signal was acquired, thereby obtaining the total energy value of the piezoelectric characteristic in the first portion of the signal. The device then averages the area of ​​the curve over the time period during which the first portion of the signal was acquired to obtain the mean energy value of the collision characteristic in the first portion of the signal, thereby obtaining the trend of change in the piezoelectric characteristic of the first portion of the signal. In some embodiments, the piezoelectric curve equation may be processed by squaring the piezoelectric curve equation.

[0101] In some embodiments, the variation trend of the piezoelectric characteristic in the first portion of the signal can be realized by formula (5):

[0102]

[0103] Among them, E aver The average value of the piezoelectric signal energy collected by the piezoelectric sensor within the time range from the current time t to the time tw (that is, the changing trend of the piezoelectric characteristics in the first part of the signal); p is the piezoelectric curve; t is the start time of signal collection, and w is the time region width of the first part of the signal.

[0104] In an embodiment of the present application, the vehicle collision recognition device can directly calculate the changing trend of the piezoelectric characteristic in the first partial signal and then determine the collision probability of the vehicle based on the changing trend, without first determining the sub-changing trend of the window signal within the first partial signal and then determining the changing trend of the piezoelectric characteristic in the first partial signal based on the sub-changing trend of the window signal. This can improve the efficiency of determining the changing trend.

[0105] In an embodiment of the present application, the vehicle collision recognition device can obtain a preset change trend and determine the collision probability of the vehicle based on the preset change trend and the change trend.

[0106] In some embodiments, if the change trend is greater than a preset change trend, the vehicle collision recognition device determines a collision probability that satisfies a preset collision condition; if the change trend is less than or equal to the preset change trend, the vehicle collision recognition device determines a collision probability that does not satisfy the preset collision condition. The preset collision condition is used to indicate a high probability of a vehicle collision.

[0107] The embodiments of the present application can determine the sub-trends of collision characteristics within each window signal; statistically process at least two of the sub-trends to obtain a variation trend for the collision characteristics within the first portion of the signal; and determine the collision probability of the vehicle based on the variation trend. Thus, by calculating the variation trend for the collision characteristics within the first portion of the signal, the variation state of the collision characteristics can be determined. The collision probability of the vehicle can then be determined based on the variation state of the collision characteristics, thereby improving the accuracy of determining the collision probability of the vehicle.

[0108] Figure 4 This embodiment of the present application provides Figure 2 The flow diagram of S204 is shown in the figure. Figure 2 S204 in the above can be realized by S401 to S404, which will be combined with Figure 4 The steps shown are explained.

[0109] In S401, filtering and transforming processing is performed on the second part of the signal to obtain a filtered transformed signal; the filtered transformed signal includes the filtered signal and at least one other filtered transformed signal except the filtered signal.

[0110] Here, the filtering and transforming process may include filtering and transforming. In some embodiments, performing filtering and transforming on the second portion of the signal to obtain a filtered and transformed signal may include: performing filtering on the second portion of the signal to obtain a filtered signal; and performing transforming on the filtered signal to obtain at least one other filtered and transformed signal.

[0111] In some embodiments, other filtered transformed signals may include: a filtered residual signal, a wavelet signal, a wavelet reconstruction signal, and a wavelet residual signal; correspondingly, transforming the filtered signal to obtain at least one other filtered transformed signal may include: obtaining a filtered residual signal based on the difference between the filtered signal and the second part of the signal; performing a wavelet transform on the filtered signal to obtain a wavelet signal; performing an inverse wavelet transform on the wavelet signal to obtain a wavelet reconstruction signal; and obtaining a wavelet residual signal based on the difference between the wavelet reconstruction signal and the wavelet signal. In this way, by performing a filtered transform on the second part of the signal, low-frequency vibration noise of the hardware itself and some high-frequency electronic noise can be removed, and the sensitivity of the signal to the environment can be reduced.

[0112] In S402 , for each other filtered transformed signal, a value range conversion is performed on the other filtered transformed signal to obtain at least one value range converted signal.

[0113] Here, by performing range conversion on the other filtered transformed signals, signals of different dimensions can be obtained, i.e., the dimensions of the range converted signals are different from those of the other filtered transformed signals. For example, when the other filtered transformed signals are time domain signals, the range converted signals can be frequency domain signals, and the range conversion in this case can be a Fourier transform; when the range converted signals are envelope domain signals, the range conversion in this case can be a Hilbert transform; and when the range converted signals are scale domain signals, the range conversion in this case can be a Mellin transform.

[0114] In some embodiments, S402 may include: performing a Fourier transform, a Hilbert transform, and a Mellin transform on each other filtered transformed signal, respectively, to obtain a frequency domain signal corresponding to the Fourier transform, an envelope domain signal corresponding to the Hilbert transform, and a scale domain signal corresponding to the Mellin transform. That is, the at least one frequency domain signal includes: a frequency domain signal obtained by Fourier transforming each other filtered transformed signal; the at least one envelope domain signal includes: an envelope domain signal obtained by Hilbert transforming each other filtered transformed signal; and the at least one scale domain signal includes: a scale domain signal obtained by Mellin transforming each other filtered transformed signal. In this case, the at least one frequency domain signal may be used as a frequency domain signal set, the at least one envelope domain signal may be used as an envelope domain signal set, and the at least one scale domain signal may be used as a scale domain signal set.

[0115] In the case where other filtered transformation signals include filtered residual signals, wavelet signals, wavelet reconstruction signals and wavelet residual signals, the above-mentioned frequency domain signal set includes: the frequency domain signal obtained by Fourier transforming the filtered residual signal, the frequency domain signal obtained by Fourier transforming the wavelet signal, the frequency domain signal obtained by Fourier transforming the wavelet reconstruction signal and the frequency domain signal obtained by Fourier transforming the wavelet residual signal; the above-mentioned envelope domain signal set includes: the envelope domain signal obtained by Hilbert transforming the filtered residual signal, the envelope domain signal obtained by Hilbert transforming the wavelet signal, the envelope domain signal obtained by Hilbert transforming the wavelet reconstruction signal and the envelope domain signal obtained by Hilbert transforming the wavelet residual signal; the above-mentioned scale domain signal set includes: the scale domain signal obtained by Mellin transforming the filtered residual signal, the scale domain signal obtained by Mellin transforming the wavelet signal, the scale domain signal obtained by Mellin transforming the wavelet reconstruction signal and the scale domain signal obtained by Mellin transforming the wavelet residual signal.

[0116] In S403 , feature extraction is performed on the filtered signal and the at least one range-converted signal to obtain a peak feature set.

[0117] In the embodiment of the present application, the vehicle collision recognition device can perform feature extraction on the filtered signal and each range conversion signal to obtain a peak feature set.

[0118] In some embodiments, when at least one value domain conversion signal includes the above-mentioned frequency domain signal set, envelope domain signal set and scale domain signal set, feature extraction can be performed on each signal in the frequency domain signal set, envelope domain signal set and scale domain signal set, and feature extraction can be performed on the filtered signal, so as to obtain a frequency domain feature set corresponding to the frequency domain signal set, an envelope domain feature set corresponding to the envelope domain signal set, a scale domain feature set corresponding to the scale domain signal set and a time domain feature set corresponding to the filtered signal, and the frequency domain feature set, envelope domain feature set, scale domain feature set and time domain feature set are used as peak feature sets.

[0119] In some embodiments, the peak feature set may include at least one of the following: maximum peak, second largest peak, third largest peak, first statistical feature, second statistical feature, energy number, curve length, peak-to-peak value, average value, root mean square value, skewness and kurtosis.

[0120] In some embodiments, the maximum peak value is used to represent the maximum amplitude of the signal to be extracted; the second largest peak value is used to represent the second largest amplitude of the signal to be extracted; and the third largest peak value is used to represent the third largest amplitude of the signal to be extracted. The signal to be extracted can be any one of a frequency domain signal set, an envelope domain signal set, a scale domain signal set, and a filtered signal.

[0121] In some embodiments, the first statistical feature is used to characterize a statistical value of peaks exceeding a first preset percentage of a maximum peak value in the signal to be extracted; the statistical value includes at least one of the following: the number of peaks exceeding the first preset percentage of the maximum peak value, the mean of the amplitudes of the peaks exceeding the first preset percentage of the maximum peak value, and the variance of the amplitudes of the peaks exceeding the first preset percentage of the maximum peak value. In some embodiments, the first preset percentage may be 20%.

[0122] In some embodiments, the second statistical feature is used to characterize a statistical value of peaks exceeding a second preset percentage of a maximum peak value in the to-be-extracted signal; the statistical value includes at least one of the following: the number of peaks exceeding the second preset percentage of the maximum peak value, the mean of the amplitudes of the peaks exceeding the second preset percentage of the maximum peak value, and the variance of the amplitudes of the peaks exceeding the second preset percentage of the maximum peak value. In some embodiments, the second preset percentage may be 80%.

[0123] In some embodiments, the energy number is used to characterize the distribution position of the waveform of the signal to be extracted, wherein the energy feature can be realized by formula (6):

[0124]

[0125] Among them, norm is the energy characteristic; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted.

[0126] In some embodiments, the curve length is used to describe the complexity of the signal. Changes in the curve length may be caused by changes in the amplitude and position of the waveform, and have good robustness to changes in the time scale. The curve length can be achieved using formula (7):

[0127]

[0128] Where cl is the curve length; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted.

[0129] In some embodiments, the peak-to-peak value is used to characterize the change in the maximum amplitude of the signal to be extracted, wherein the peak-to-peak value can be realized by formula (8):

[0130] mm=max 1th -min formula(8)

[0131] Among them, mm is the peak-to-peak value; max 1th is the highest value of the signal to be extracted; min is the lowest value of the signal to be extracted.

[0132] In some embodiments, the average value is used to characterize the expected waveform of the signal to be extracted, wherein the average value can be implemented by formula (9):

[0133]

[0134] Where μ is the average value; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted.

[0135] In some embodiments, the root mean square value is used to characterize the distribution of the waveform of the signal to be extracted. The root mean square value can be realized by formula (10):

[0136]

[0137] Where σ is the root mean square value; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted.

[0138] In some embodiments, the skewness is used to characterize the degree of skewness of the waveform of the signal to be extracted. The skewness can be implemented by formula (11):

[0139]

[0140] Where k is the skewness; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted; μ is the mean value of all sampling time windows in the signal to be extracted; σ is the root mean square value of the signal to be extracted.

[0141] In some embodiments, the kurtosis is used to characterize the peak sharpness of the waveform of the signal to be extracted. The kurtosis can be implemented by formula (12):

[0142]

[0143] Where ku is the kurtosis; x i is the mean value of the signal to be extracted in the i-th sampling time window; n is the total number of all sampling time windows in the signal to be extracted; μ is the mean value of all sampling time windows in the signal to be extracted.

[0144] In S404 , the peak feature set is input into the trained deep learning model to obtain a vehicle collision recognition result.

[0145] In an embodiment of the present application, the vehicle collision recognition device can input the peak feature set into a trained deep learning model, and the trained deep learning model outputs the vehicle collision recognition result.

[0146] In some embodiments, the collision recognition result may include a collision recognition result indicating that the vehicle is not damaged, a collision recognition result indicating that the vehicle is slightly impacted, a collision recognition result indicating that the vehicle is moderately impacted, and a collision recognition result indicating that the vehicle is severely impacted.

[0147] In some embodiments, the deep learning model is mainly composed of an encoder-decoder architecture. The encoder is responsible for further high-dimensional representation of the features and extracting the features, and the decoder is responsible for mapping the high-dimensional features to the damage results. Mature convolutional neural networks (CNN) or recurrent neural networks (RNN) such as VGG (Visual Geometry Group) and deep residual network (ResNet) can be selected.

[0148] In some embodiments, the training process of the deep learning model includes: collecting test signals for scenarios that need to be triggered during vehicle parking and scenarios that should not be triggered. For scenarios that should not be triggered, the sample label is recorded as 0 - not damaged, such as rainy days, snowy days, vibrations caused by the passage of large trucks, honking, closing doors, opening windows, and other scenarios that do not need to be triggered; for scenarios that should be triggered, tests can be conducted according to actual conditions, such as hitting the vehicle body with a ball, scratching the vehicle body with a hard object, scratching between vehicles, hitting pedestrians, hitting vehicles, etc. According to the damage to the vehicle, the sample labels can be divided into 1 - slight impact, 2 - moderate impact, and 3 - severe impact. The training samples with sample labels are then input into the deep learning model, and the deep learning model is adjusted based on the output results and sample labels. When the number of iterations meets the preset number or the difference between the output result and the sample label is within the preset difference, the training of the deep learning model is completed.

[0149] The embodiment of the present application can perform filtering and transformation processing on the second part of the signal to obtain a filtered signal and at least one other filtered and transformed signal; this can remove the low-frequency vibration noise of the hardware itself and some high-frequency electronic noise. Then, each other filtered and transformed signal is subjected to value range conversion to obtain at least one value range conversion signal; this can obtain signals of different dimensions. Feature extraction is performed on the filtered signal and at least one value range conversion signal respectively to obtain a peak feature set; the collision recognition result of the vehicle is determined based on the peak feature set. In this way, feature extraction is performed on signals of different dimensions to obtain typical features of the waveform, and finally the collision recognition result of the vehicle is determined based on the peak feature set, which can improve the accuracy of determining the collision recognition result of the vehicle.

[0150] Figure 5This is a flow chart of the vehicle collision identification method provided by the embodiment of the present application, based on Figure 1 , Figure 1 After S104 in the above method, the above method may further include S501 to S502, combining Figure 5 The steps shown are explained.

[0151] In S501, when the collision recognition result of the vehicle indicates that the vehicle has collided, a first control instruction is sent to the video acquisition device of the vehicle to instruct the video acquisition device to acquire video data, and a second control instruction is sent to the position acquisition device of the vehicle to instruct the position acquisition device to acquire position information of the vehicle.

[0152] In S502, video data and location information are received, and the video data, location information and collision recognition results are sent to the server; the server is used to send the video data, location information and collision recognition results to the client.

[0153] In an embodiment of the present application, the vehicle collision recognition device can, when determining that the collision recognition result indicates that the vehicle has collided, send a first control instruction to the vehicle's video acquisition device to instruct the video acquisition device to collect video data, and send a second control instruction to the vehicle's position acquisition device to instruct the position acquisition device to collect the vehicle's position information, so as to collect video information of the actual vehicle collision and the position of the vehicle at the time of the collision; then, the vehicle collision recognition device can send the received video data, position information, and determined collision recognition result to the server through the vehicle's network module, so that the server sends the video data, position information, and collision recognition result to the client. In this way, the user can determine the current vehicle's collision status in a timely manner through the client.

[0154] In some embodiments, after receiving the video data, location information, and collision recognition results, the client sends feedback information to the server to indicate whether the collision recognition results are correct. If the feedback information indicates that the collision recognition results are incorrect, the server can retrain the deep learning model in the vehicle collision recognition device based on the feedback information to improve the accuracy of the deep learning model in identifying collisions.

[0155] Figure 6 This is a schematic diagram of the implementation framework of the vehicle collision recognition method provided in the embodiment of the present application. Figure 6 In the embodiment, the implementation framework of the vehicle collision identification method includes: a data collection module 601, an identification module 602, a network module 603, a server module 604 and a client module 605; wherein,

[0156] The data collection module 601 includes an acceleration sensor module 611 (equivalent to the acceleration sensor in the above embodiment), an acceleration sensor filtering module 612, a speed sensing module 613, a piezoelectric sensor module 614 (equivalent to the piezoelectric sensor in the above embodiment), a piezoelectric sensor filtering and amplification module 615, a camera module 616 and a GPS module 617.

[0157] The acceleration sensor module 611 collects the time domain signal of the vehicle's collision acceleration that changes with time (equivalent to the acceleration signal in the above embodiment), the piezoelectric sensor module 614 collects the piezoelectric signal that changes with time when the vehicle vibrates (equivalent to the piezoelectric signal in the above embodiment), the camera module 616 collects image information around the vehicle, the GPS module 617 obtains the vehicle's location information in real time, the speed sensing module 613 collects the current vehicle's driving speed, the acceleration sensor filtering module 612 filters the acceleration signal to remove the hardware's own low-frequency vibration noise and some high-frequency electronic noise, and the piezoelectric sensor filtering and amplification module 615 filters the piezoelectric signal and amplifies the signal to reduce the impact of the low-frequency signal and reduce the signal's sensitivity to the environment.

[0158] The recognition module 602 includes a discrimination module 621, a aver Value calculation module 622, E aver Value calculation module 623, signal window acquisition module 624, signal characteristic value calculation module 625, deep learning module 626 and damage recording module 627; wherein,

[0159] The k aver The value calculation module 622 is used to calculate the acceleration change trend, that is, the average value k of the k values ​​of consecutive m time steps. aver In some embodiments, k aver This can be achieved by formula (13):

[0160]

[0161] Among them, k i is the k value of the i-th time step, k i This can be achieved by formula (14):

[0162]

[0163] Among them, a i is the i-th value of the synthetic acceleration time curve, a i This can be achieved by formula (15):

[0164]

[0165] Among them, aix 、a iy 、a iz are the accelerations in the X, Y, and Z directions collected by the acceleration sensor at the i-th time step.

[0166] In some embodiments, E aver The value calculation module 623 can be implemented by formula (16):

[0167]

[0168] Where p is the piezoelectric curve, t is the start time of signal acquisition, and w is the width of the preset time window, which is generally an integer multiple of the sampling period. aver Represents the average value of the piezoelectric signal energy collected by the piezoelectric sensor within the time range from the current time t to the time tw.

[0169] The judging module 621 is used to determine the vehicle speed V,k aver , E aver Preliminary judgment is made on whether a collision may have occurred, so as to make further judgment. thres , the vehicle is considered to be in dynamic driving. When a collision occurs, the acceleration value is generally large. Select the acceleration k aver As a trigger parameter for further judgment, when k aver >k thres , to conduct further collision identification; when V <V thres , then the vehicle is considered to be in a quasi-static environment, and the energy value of the piezoelectric signal E aver As a trigger parameter for further judgment, when E aver >E thres , for further collision identification.

[0170] The signal window acquisition module 624 acquires the signal of the time window W, and when k aver >k thres or E aver >E thres , then the signal is collected according to the current time step, and the starting time of signal collection is T start =tW, t is the current time, and W can be 50ms.

[0171] The signal eigenvalue calculation module 625 is used to perform range conversion on the signal collected by the signal window acquisition module 624 and calculate the signal eigenvalues. The performance of deep learning is highly dependent on whether key features are found. It is generally difficult to find eigenvalues ​​that are sensitive to scene recognition and robust to environmental ambiguity. This module provides a framework for finding a large number of potential features.

[0172] In some embodiments, the wavelet transform is a time-scale analysis method for signals. It has excellent local signal characterization capabilities, with high temporal resolution and low frequency resolution in low-frequency regions, making it suitable for analyzing non-stationary signals and extracting their local features. The wavelet transform decomposes a waveform into a set of orthogonal function bases. The mother wavelet is derived by translation and scaling of a mother wavelet, which can be a 4th-order Symlets mother wavelet.

[0173] Take the calculation of acceleration signal characteristic value as an example:

[0174] First, the signal eigenvalue calculation module 625 obtains the filtered residual signal S2 (equivalent to the filtered residual signal in the above embodiment) based on the difference between the original signal S0 (equivalent to the second part of the signal in the above embodiment) and the filtered signal S1 (equivalent to the filtered signal in the above embodiment), and performs a wavelet transform on the filtered time domain signal S1 to obtain a wavelet coefficient signal S3 (equivalent to the wavelet signal in the above embodiment), and then performs an inverse wavelet transform to obtain a wavelet reconstruction signal S4 (equivalent to the wavelet reconstruction signal in the above embodiment), and calculates the wavelet residual signal S5 (equivalent to the wavelet residual signal in the above embodiment).

[0175] Then, the signal eigenvalue calculation module 625 performs Fourier transform, Hilbert transform, and Mellin transform on S1, S2, S3, S4, and S5, respectively. The Fourier transform can obtain the signal characteristics in the frequency domain (frequency domain). The Hilbert transform converts the amplitude of the signal into the signal envelope of the original signal (envelope domain). The Mellin transform is scale-invariant, so it can obtain more environmentally insensitive eigenvalues ​​in the scale domain.

[0176] Finally, the signal eigenvalue calculation module 625 extracts signal eigenvalues ​​from the time domain signal, frequency domain signal, envelope domain signal and scale domain signal respectively, which are defined as the time domain signal eigenvalue, frequency domain signal eigenvalue, envelope domain signal eigenvalue and scale domain signal eigenvalue, collectively referred to as the parent signal eigenvalue, respectively denoted as Each type of parent signal feature value contains numerous sub-signal feature values, which may include but are not limited to:

[0177] The peak-related eigenvalue reflects the typical characteristics of the waveform. Different collisions have different peaks. It is a theoretically better eigenvalue. This patent selects the first peak (maximum peak) max 1th , the second peak max 2th , the third peak max 3th , the number of peaks exceeding 20% ​​of the first peak 20per , mean μ 20per and variance σ 20per , the number of peaks exceeding 80% of the first peak80per , mean μ 80per and variance σ 80per ;

[0178] Energy number Reflects the distribution position of the waveform;

[0179] Curve length It helps to describe the complexity of the signal. The change of the curve length may be caused by the change of the amplitude and position of the waveform. It has good robustness to the change of time scale.

[0180] Peak-to-peak value mm=max 1th -min, reflects the change in maximum amplitude;

[0181] Mean (first moment) expectations of response waveforms;

[0182] RMS value (second moment) distribution of response waveforms;

[0183] Skewness (third moment) Reflects the degree of waveform skewness;

[0184] Kurtosis (fourth moment) Reflects the sharpness of the peak of the waveform.

[0185] In the above formula, n is the number of signal samples, and Xi is the value of the parent signal feature at the i-th time step.

[0186] According to the above method, the extracted acceleration signal characteristic value is recorded as Similarly, the characteristic value of the piezoelectric signal can be obtained as

[0187] At this point, the acceleration signal characteristic value F is obtained a and the piezoelectric signal characteristic value F p , and F a and F p As input to deep learning algorithms.

[0188] The deep learning module 626 is an offline, pre-trained deep learning model that is used to predict whether a vehicle will collide and the damage results after a collision based on the collected signals. The construction process of the deep learning model includes:

[0189] S1 Test sample collection: After the acceleration sensor and piezoelectric sensor are installed on the vehicle, samples are collected. Specifically, test signals are collected for scenes that need to be triggered during vehicle parking and scenes that should not be triggered. For scenes that should not be triggered, the sample label is recorded as 0-not damaged, such as rainy days, snow, vibrations caused by the passage of large trucks, honking, closing doors, opening windows, and other scenes that do not need to be triggered; for scenes that should be triggered, tests can be carried out according to actual conditions, such as hitting the vehicle body with a ball, scratching the vehicle body with a hard object, scratching between vehicles, hitting pedestrians, hitting vehicles, etc. According to the damage to the vehicle, the sample labels can be divided into 1-slightly hit, 2-medium hit, and 3-severely hit. After completing the collection of acceleration sensor signals and piezoelectric sensor signals for all the above scenarios. The output (label) is a sample set of damage results (0, 1, 2, 3). It is worth noting that the output of the sample is not limited to the four results described, and can also be more detailed damage results.

[0190] S2 deep learning model construction: It mainly consists of an encoder-decoder architecture. The encoder is responsible for further high-dimensional representation of features and feature extraction, and the decoder is responsible for mapping high-dimensional features to damage results. Mature convolutional neural networks (CNNs) or recurrent neural networks (RNNs) such as VGG and Resnet can be selected.

[0191] S3 deep learning model training: Input training samples with sample labels into the deep learning model, adjust the deep learning model based on the output results and sample labels, and complete the training of the deep learning model when the number of iterations meets the preset number or the difference between the output results and the sample labels is within the preset difference.

[0192] S4 Deep learning model deployment: Deploy the trained deep learning model in the recognition module 602.

[0193] After the deep learning module 626 determines that the vehicle has collided and determines the damage results after the collision, the damage results are sent to the damage recording module 627. At the same time, the camera module 616 collects image information around the vehicle, and the GPS module 617 obtains the vehicle's location information in real time and sends the image information and location information to the damage recording module 627. The damage recording module 627 sends the damage results, image information and location information to the server module 604 through the network module 603.

[0194] The server module 604 is configured to send the damage results, image information, and location information to the client module 605. The client module 605 is configured to send user feedback on the damage results to the server module 604. When the server module 604 receives customer feedback confirming that the actual damage results are inconsistent with the results predicted by the recognition module 602, the server module 604 records the sample, retrains the deep learning module 626 in the recognition module 602, and updates the deep learning module 626 to improve prediction accuracy.

[0195] Figure 7 This is a scene framework diagram of the vehicle collision recognition method provided in the embodiment of the present application, such as Figure 7 As shown, the scenario framework of the vehicle collision recognition method includes: a vehicle-side recognition module 701, a network module 702, a server module 703, and a client-side module 704. The vehicle-side recognition module 701 is deployed in the vehicle collision recognition device and is used to obtain a vehicle collision recognition result using the vehicle collision recognition method described in the above embodiment. If the vehicle-side recognition module determines that the vehicle has suffered a minor collision, a moderate collision, or a severe collision, it triggers the interior and exterior cameras to begin recording, with a customizable recording duration. After triggering the cameras to record and completing the recording, the network module 702 is used to upload the specific damage results, video data, and GPS information to the server module 703. The server module 703 stores each information uploaded by the network module 702 and immediately notifies the client module 704 upon receipt. The client module 704 is an application pre-installed on the client's mobile phone or computer. After receiving the vehicle damage information, the client can verify the accuracy of the reported information based on the actual situation and feedback the results to the server 703 via the client module 704. When the server side 703 receives customer feedback and confirms that the actual damage result is inconsistent with the result predicted by the vehicle-side recognition module 701, the sample is recorded, the deep learning module is retrained, and the deep learning module is updated to improve the prediction accuracy.

[0196] The embodiment of the present application provides a vehicle collision recognition device, Figure 8 A schematic diagram of the structure of a vehicle collision recognition device 800 provided in an embodiment of the present application is shown as follows: Figure 8 As shown, the apparatus includes: an acquisition unit 801, a first determination unit 802, a second determination unit 803, and an identification unit 804, wherein:

[0197] An acquisition unit 801 is used to acquire an acceleration signal and a piezoelectric signal of a vehicle;

[0198] A first determining unit 802 is configured to determine a driving speed of the vehicle when acquiring the acceleration signal and the piezoelectric signal;

[0199] A second determining unit 803 is configured to determine a target signal from the acceleration signal and the piezoelectric signal based on the driving speed;

[0200] The identification unit 804 is configured to determine a collision identification result of the vehicle based on the target signal.

[0201] In some embodiments, the first collision signal is the acceleration signal; the second collision signal is the piezoelectric signal; the second determination unit 803 is further used to determine the acceleration signal as the target signal when the driving speed is greater than a preset driving speed; and to determine the piezoelectric signal as the target signal when the driving speed is less than or equal to the preset driving speed.

[0202] In some embodiments, the identification unit 804 is further used to determine a changing trend of the collision characteristics of the first part of the signal in the target signal based on the first part of the signal; determine the collision probability of the vehicle based on the changing trend of the collision characteristics of the first part of the signal; obtain a second part of the signal in the target signal when the collision probability meets a preset collision condition; and determine a collision identification result of the vehicle based on the second part of the signal.

[0203] In some embodiments, the first part of the signal includes at least two continuous window signals; the identification unit 804 is further used to, for each of the window signals, perform statistical processing on at least two of the sub-change trends based on the collision characteristics of the window signal to obtain the change trend of the collision characteristics of the first part of the signal.

[0204] In some embodiments, when the target signal is the acceleration signal, the collision feature includes an acceleration feature; the identification unit 804 is further used to obtain at least two acceleration features on the window signal for each window signal; based on the at least two acceleration features and the time interval between two adjacent acceleration features, determine the sub-change trend of the acceleration feature of the window signal.

[0205] In some embodiments, when the target signal is the piezoelectric signal, the collision feature includes a piezoelectric feature; the identification unit 804 is further used to determine the total energy value of the piezoelectric feature in the first part of the signal; based on the total energy value, determine the change trend of the piezoelectric feature in the first part of the signal.

[0206] In some embodiments, the identification unit 804 is further configured to use the difference between the current time information and a preset time period as the start acquisition time, and the current time information as the end acquisition time; and obtain the second part of the signal from the target signal based on the start acquisition time and the end acquisition time.

[0207] In some embodiments, the identification unit 804 is further used to perform filtering and transformation processing on the second part of the signal to obtain a filtered transformation signal; the filtered transformation signal includes a filtered signal and at least one other filtered transformation signal other than the filtered signal; for each of the other filtered transformation signals, the other filtered transformation signal is converted into a value range to obtain at least one value range conversion signal; feature extraction is performed on the filtered signal and at least one of the value range conversion signals respectively to obtain a peak feature set; the peak feature set is input into the trained deep learning model to obtain a collision recognition result of the vehicle.

[0208] In some embodiments, the vehicle collision identification device 800 also includes a control unit and a sending unit; the control unit is used to send a first control instruction to the video acquisition device of the vehicle for instructing the video acquisition device to collect video data, and to send a second control instruction to the position acquisition device of the vehicle for instructing the position acquisition device to collect the position information of the vehicle when the collision identification result of the vehicle indicates that the vehicle has collided; the sending unit is used to receive the video data and the position information, and send the video data, the position information and the collision identification result to the server; the server is used to send the video data, the position information and the collision identification result to the client.

[0209] The description of the above device embodiment is similar to the description of the above method embodiment and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to perform the methods described in the above method embodiments. For technical details not disclosed in the device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0210] It should be noted that, in the embodiment of the present application, if the above-mentioned vehicle collision identification method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific hardware, software or firmware, or any combination of hardware, software and firmware.

[0211] An embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0212] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements some or all of the steps in the above method. The computer-readable storage medium may be transient or non-transient.

[0213] An embodiment of the present application provides a computer program, including computer-readable code. When the computer-readable code is run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0214] An embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, implements some or all of the steps in the above method. The computer program product can be implemented specifically by hardware, software, or a combination thereof. In some embodiments, the computer program product is embodied as a computer storage medium. In other embodiments, the computer program product is embodied as a software product, such as a software development kit (SDK), etc.

[0215] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between the various embodiments, and their similarities or similarities can be referenced to each other. The descriptions of the above device, storage medium, computer program, and computer program product embodiments are similar to the descriptions of the above method embodiments and have similar beneficial effects as the method embodiments. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the description of the method embodiments of this application for understanding.

[0216] The present invention provides a vehicle collision recognition device. Figure 9 The schematic diagram of the structure of the vehicle collision recognition device 900 provided in the embodiment of the present application is as follows: Figure 9 As shown, the device includes: a processor 901, a communication interface 902 and a memory 903, wherein:

[0217] The processor 901 generally controls the overall operation of the computer device 900, and the overall operation may be to implement the vehicle collision recognition method provided in the embodiment of the present application, for example, Figures 1 to 5 The method shown.

[0218] The communication interface 902 enables the computer device to communicate with other terminals or servers through a network.

[0219] The memory 903 is configured to store instructions and applications executable by the processor 901. It can also cache data to be processed or processed by the processor 901 and various modules in the computer device 900 (e.g., image data, audio data, voice communication data, and video communication data). This can be implemented using flash memory (FLASH) or random access memory (RAM). Data can be transmitted between the processor 901, the communication interface 902, and the memory 903 via a bus 904.

[0220] The present invention provides a computer program product or computer program, which includes computer instructions stored in a readable storage medium. A processor of a computer device reads the computer instructions from the readable storage medium and executes the computer instructions, causing the computer device to perform the vehicle collision identification method described above in the present invention.

[0221] The embodiment of the present application provides a readable storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the vehicle collision recognition method provided by the embodiment of the present application, for example, Figures 1 to 5 The method shown.

[0222] It should be noted that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0223] The processor may be at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, and a microprocessor. It is understood that the electronic device that implements the functions of the processor may also be other electronic devices, which are not specifically limited in the embodiments of the present application.

[0224] The above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0225] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned steps / processes does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0226] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0227] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0228] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0229] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0230] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

[0231] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0232] The above is only an implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A vehicle collision recognition method, characterized in that: include: Obtaining the vehicle's acceleration signal and piezoelectric signal; determining a travel speed of the vehicle when acquiring the acceleration signal and the piezoelectric signal; When the driving speed is greater than a preset driving speed, determining the acceleration signal as a target signal; When the driving speed is less than or equal to a preset driving speed, determining the piezoelectric signal as a target signal; A collision recognition result of the vehicle is determined based on the target signal.

2. The method according to claim 1, characterized in that Determining a collision recognition result of the vehicle based on the target signal includes: determining, based on a first portion of the target signal, a change trend of a collision characteristic of the first portion of the target signal; determining a collision probability of the vehicle based on a change trend of the collision characteristics of the first portion of signals; When the collision probability satisfies a preset collision condition, obtaining a second portion of the signal from the target signal; A collision recognition result of the vehicle is determined based on the second part of the signal.

3. The method according to claim 2, characterized in that The first portion of the signal includes at least two consecutive window signals; The determining, based on the first portion of the target signal, a change trend of the collision characteristic of the first portion of the target signal includes: For each of the window signals, determining a sub-change trend of the collision feature of the window signal; Statistical processing is performed on at least two of the sub-change trends to obtain a change trend of the collision characteristics of the first part of the signal.

4. The method according to claim 3, characterized in that In the case where the target signal is the acceleration signal, the collision feature includes an acceleration feature; The step of determining, for each window signal, a sub-change trend of a collision characteristic of the window signal includes: For each of the window signals, obtaining at least two acceleration features on the window signal; Based on at least two of the acceleration features and a time interval between two adjacent acceleration features, a sub-change trend of the acceleration feature of the window signal is determined.

5. The method according to claim 2, characterized in that In the case where the target signal is the piezoelectric signal, the collision feature includes a piezoelectric feature; The determining, based on the first portion of the target signal, a change trend of the collision characteristic of the first portion of the target signal includes: determining a total energy value of the piezoelectric feature in the first portion of the signal; Based on the total energy value, a change trend of the piezoelectric characteristic in the first portion of the signal is determined.

6. The method according to claim 2, characterized in that The acquiring the second portion of the signal from the target signal includes: The difference between the current time information and the preset time period is used as the start time of collection, and the current time information is used as the end time of collection; The second portion of the signal is acquired from the target signal based on the start acquisition time and the end acquisition time.

7. The method according to claim 2, characterized in that The determining of a collision recognition result of the vehicle based on the second portion of the signal includes: Performing filtering and transformation processing on the second part of the signal to obtain a filtered transformed signal; the filtered transformed signal includes a filtered signal and at least one other filtered transformed signal other than the filtered signal; For each of the other filtered transformed signals, performing a value range conversion on the other filtered transformed signal to obtain at least one value range converted signal; Performing feature extraction on the filtered signal and at least one of the range-converted signals to obtain a peak feature set; The peak feature set is input into the trained deep learning model to obtain the collision recognition result of the vehicle.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: When the collision recognition result of the vehicle indicates that the vehicle has collided, a first control instruction is sent to a video acquisition device of the vehicle to instruct the video acquisition device to acquire video data, and a second control instruction is sent to a position acquisition device of the vehicle to instruct the position acquisition device to acquire position information of the vehicle; The video data and the location information are received, and the video data, the location information and the collision recognition result are sent to a server; the server is used to send the video data, the location information and the collision recognition result to a client.

9. A vehicle collision recognition device, characterized in that: include: An acquisition unit, used to acquire an acceleration signal and a piezoelectric signal of the vehicle; a first determining unit, configured to determine a travel speed of the vehicle when acquiring the acceleration signal and the piezoelectric signal; a second selection unit, configured to determine the acceleration signal as a target signal when the driving speed is greater than a preset driving speed; When the driving speed is less than or equal to a preset driving speed, determining the piezoelectric signal as a target signal; The recognition unit is configured to determine a collision recognition result of the vehicle based on the target signal.

Images