Vehicle-mounted collision detection method, system and platform based on multi-dimensional feature fusion

Abstract

The invention discloses a vehicle-mounted collision detection method, system and platform based on multi-dimensional feature fusion. First data corresponding to a vehicle are generated and acquired, and the first data are processed in a segmented manner in combination with a fixed time window; wherein the first data is vehicle three-axis acceleration data; generating and acquiring second data corresponding to the first data, and judging and generating third data corresponding to vehicle-mounted collision based on multi-dimensional feature fusion; wherein the second data is a preset dynamic threshold value; the third data is vehicle-mounted collision event detection data, and the system and the platform corresponding to the method realize high-precision collision judgment by fusing acceleration, Jerk and wavelet packet energy analysis. According to the technical scheme, the method comprises the steps of data collection, multiple threshold value judgment (Jerk, proportion, acceleration and frequency domain energy) and dynamic adjustment, the false alarm rate and the omission ratio are remarkably reduced, and the method is suitable for intelligent network connection automobile safety monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent connected vehicles, and specifically relates to a vehicle-mounted collision detection method, system and platform based on multi-dimensional feature fusion. Background Art

[0002] With the development of intelligent connected vehicles and autonomous driving technologies, real-time vehicle status monitoring and emergency event reporting have become key requirements. Traditional collision detection methods primarily rely on a single acceleration threshold. For example, when the acceleration detected by the IMU (Inertial Measurement Unit) exceeds a preset threshold, a collision is detected. However, when a vehicle accelerates or decelerates suddenly, or when it passes over speed bumps or potholes, the acceleration may exceed the threshold, leading to misjudgment. Sensor installation positions, angles, and chassis suspension characteristics vary from vehicle to vehicle, making a single threshold difficult to adapt to all scenarios. Moreover, minor collisions or scrapes may be ignored because the acceleration does not reach the threshold. After-installed onboard monitoring equipment not only collects video images and vehicle condition data (such as speed and steering angle) during driving, but also needs to monitor the vehicle in real time based on the IMU sensor to determine if a collision has occurred. Once a collision event is detected, the equipment needs to separately analyze and record the collision data and vehicle data.

[0003] Furthermore, traditional collision detection often relies on an acceleration threshold method: a single acceleration threshold is set, and if the acceleration detected by the sensor exceeds this threshold, a collision is considered to have occurred. However, in practice, this method can result in a high false alarm rate due to vehicle vibration, sudden acceleration and deceleration, and driving over potholes or speed bumps. Furthermore, the location and angle of sensor mounting equipment vary across vehicles and operators, so a single threshold often cannot meet the needs of all vehicles, leading to missed detections.

[0004] Therefore, in view of the above technical problems and defects such as high false alarm rate, poor adaptability, missed detection risk, and 3. complex environmental interference, it is urgent to design and develop a collision detection method with high adaptability and low false alarm rate, that is, a vehicle-mounted collision detection method, system and platform based on multi-dimensional feature fusion. Summary of the Invention

[0005] In order to overcome the shortcomings and difficulties of the above-mentioned existing technologies, the purpose of the present invention is to provide a vehicle-mounted collision detection method, system and platform based on multi-dimensional feature fusion. By combining acceleration, Jerk and wavelet packet energy analysis, high-precision collision detection is achieved, and dynamic threshold adjustment is supported to adapt to different vehicles and installation conditions.

[0006] The first purpose of the present invention is to provide a vehicle-mounted collision detection method based on multi-dimensional feature fusion; the second purpose of the present invention is to provide a vehicle-mounted collision detection system based on multi-dimensional feature fusion; the third purpose of the present invention is to provide a vehicle-mounted collision detection platform based on multi-dimensional feature fusion.

[0007] The first object of the present invention is achieved in that the method comprises the steps of:

[0008] Generate and obtain first data corresponding to the vehicle, and process the first data in segments based on a fixed time window; wherein the first data is three-axis acceleration data of the vehicle;

[0009] Generate and obtain second data corresponding to the first data, and based on multi-dimensional feature fusion, determine and generate third data corresponding to the vehicle collision; wherein, the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

[0010] Furthermore, the generating and acquiring of first data corresponding to the vehicle and processing the first data in segments in combination with a fixed time window further includes:

[0011] The vehicle's three-axis acceleration data is collected in real time through the on-board IMU sensor, with a sampling frequency of no less than 200Hz;

[0012] The collected acceleration data is segmented and processed according to fixed time windows.

[0013] Furthermore, the generating and acquiring of second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0014] Calculating and generating fourth data corresponding to the first data, and executing the next step if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data; wherein the fourth data is the rate of change of acceleration of the three axes;

[0015] Calculate and generate ratio data of the Z axis and the XY axis corresponding to the fourth data, and execute the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold;

[0016] Calculate and generate the maximum acceleration value of the XY axis. When it exceeds the dynamically adjusted acceleration threshold, it is directly determined as a collision;

[0017] Perform wavelet packet decomposition on the XY axis acceleration data and calculate the total energy of the 0-6Hz external frequency band. If it exceeds the threshold, proceed to the next step of judgment.

[0018] The standard deviation to mean ratio of the high-frequency energy distribution is calculated, and if it exceeds the distribution threshold, it is determined to be a collision event.

[0019] Furthermore, the generating and acquiring of second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0020] Set the generated basic threshold data. For rigidly mounted equipment, the basic threshold is set to 500m / s. 3 ;

[0021] For equipment that is not rigidly installed or on vehicles with strong energy absorption, the threshold processing is lowered; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are reduced proportionally.

[0022] Furthermore, the generating and acquiring of second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0023] Detect the deviation between the Z-axis acceleration and the standard gravity acceleration, calculate the rotation matrix and correct the original three-axis acceleration;

[0024] When the tilt angle exceeds the threshold, a force correction coefficient is introduced for the XY axis acceleration.

[0025] Furthermore, the generating and acquiring of second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0026] Perform wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; set the distribution threshold to 38, and determine a collision when the ratio of the energy distribution standard deviation to the mean exceeds this value;

[0027] When a collision is detected, the system automatically uploads 90 seconds of video data before and after the collision, and simultaneously uploads IMU raw data and vehicle operating status data. The reported data includes vehicle speed, steering angle, gear position, and positioning coordinate information.

[0028] The second object of the present invention is achieved as follows: the system is used to implement the vehicle-mounted collision detection method based on multi-dimensional feature fusion; the system includes:

[0029] A first data generating unit is configured to generate and obtain first data corresponding to the vehicle, and process the first data in segments in combination with a fixed time window; wherein the first data is three-axis acceleration data of the vehicle;

[0030] The second data generation unit is used to generate and obtain second data corresponding to the first data, and based on multi-dimensional feature fusion, determine and generate third data corresponding to the vehicle collision; wherein, the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

[0031] Furthermore, the first data generating unit further includes:

[0032] The first processing module is used to collect the vehicle's three-axis acceleration data in real time through the vehicle-mounted IMU sensor; wherein the sampling frequency is not less than 200 Hz;

[0033] The second processing module is used to process the collected acceleration data in segments according to fixed time windows;

[0034] And / or, the second data generating unit further includes:

[0035] a first generating module, configured to calculate and generate fourth data corresponding to the first data, and execute the next step if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data; wherein the fourth data is a rate of change of acceleration of the three axes;

[0036] a second generating module, configured to calculate and generate ratio data of the Z axis and the XY axis corresponding to the fourth data, and execute the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold;

[0037] The third generation module is used to calculate and generate the maximum acceleration value of the XY axis, and directly determine it as a collision when it exceeds the dynamically adjusted acceleration threshold;

[0038] The third processing module is used to perform wavelet packet decomposition on the XY axis acceleration data and calculate the total energy of the 0-6Hz external frequency band. If it exceeds the threshold, it will proceed to the next step of judgment;

[0039] The first determination module is used to calculate the standard deviation and mean ratio of the high-frequency energy distribution, and determine it as a collision event if it exceeds the distribution threshold.

[0040] Furthermore, the second data generating unit further includes:

[0041] The fourth generation module is used to set and generate basic threshold data, wherein the basic threshold for rigidly installed equipment is set to 500m / s 3 ;

[0042] The fifth generation module is used to lower the threshold value for devices that are not rigidly mounted or have strong energy absorption on the vehicle; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are proportionally lowered;

[0043] The fourth processing module is used to detect the deviation between the Z-axis acceleration and the standard gravity acceleration, calculate the rotation matrix and correct the original three-axis acceleration;

[0044] a fifth processing module, configured to introduce a force correction coefficient into the XY axis acceleration when the tilt angle exceeds a threshold;

[0045] The sixth processing module is used to perform wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; wherein the distribution threshold is set to 38, and a collision is determined when the ratio of the energy distribution standard deviation to the mean exceeds this value;

[0046] The seventh processing module is used to automatically upload 90 seconds of video data before and after the collision when a collision is determined to have occurred; and simultaneously upload the IMU raw data and vehicle operating status data; among which the reported data includes vehicle speed, steering angle, gear position and positioning coordinate information.

[0047] The third object of the present invention is achieved as follows: it includes a processor, a memory and a vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion; wherein the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is executed by the processor, the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is stored in the memory, and the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion implements the vehicle-mounted collision detection method based on multi-dimensional feature fusion.

[0048] The present invention generates and obtains first data corresponding to a vehicle through a method, and processes the first data in segments in combination with a fixed time window; wherein the first data is the vehicle's three-axis acceleration data; generates and obtains second data corresponding to the first data, and determines and generates third data corresponding to a vehicle collision based on multi-dimensional feature fusion; wherein the second data is a preset dynamic threshold; the third data is vehicle collision event detection data, as well as a system and platform corresponding to the method, which achieves high-precision collision determination by fusing acceleration, Jerk, and wavelet packet energy analysis. The technical solution includes data acquisition, multiple threshold determination (Jerk, ratio, acceleration, frequency domain energy), and dynamic adjustment, significantly reducing false alarm and missed detection rates, and is suitable for safety monitoring of intelligent connected vehicles.

[0049] In other words, the solution of the present invention adapts to different installation conditions and vehicle types through dynamic threshold adjustment and tilt correction. Combining Jerk ratio and wavelet packet energy analysis, it effectively filters out interference such as road bumps. Frequency domain analysis can capture minor collisions, reducing the missed detection rate. Furthermore, the solution of the present invention has low algorithm complexity, making it suitable for deployment on vehicle terminals. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 This is a flow chart of a vehicle-mounted collision detection method based on multi-dimensional feature fusion according to the present invention;

[0052] Figure 2 This is a schematic diagram of the overall flow of a collision detection method according to an embodiment of the present invention, which is a vehicle-mounted collision detection method based on multi-dimensional feature fusion;

[0053] Figure 3 This is a schematic diagram of an example of the time domain of IMU data during a collision in an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion of the present invention;

[0054] Figure 4 This is a schematic diagram of an example of the frequency domain of IMU data in a simulated collision in an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion according to the present invention;

[0055] Figure 5 This is a schematic diagram of an example of IMU data during a collision in accordance with an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion according to the present invention;

[0056] Figure 6 This is a schematic diagram of window data of a collision detection embodiment of an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion of the present invention;

[0057] Figure 7 This is a screenshot diagram of an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion according to the present invention, in which a collision event is reported on a platform;

[0058] Figure 8 This is a schematic diagram of a screenshot of a platform preview collision detection embodiment of an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion of the present invention;

[0059] Figure 9 This is a second schematic diagram of a screenshot of a platform preview collision detection embodiment of an embodiment of a vehicle-mounted collision detection method based on multi-dimensional feature fusion according to the present invention;

[0060] Figure 10 This is a schematic diagram of the architecture of a vehicle-mounted collision detection system based on multi-dimensional feature fusion according to the present invention;

[0061] Figure 11 This is a schematic diagram of the architecture of an on-board collision detection platform based on multi-dimensional feature fusion in the present invention. DETAILED DESCRIPTION

[0062] In order to better understand the purpose, technical solutions and advantages of the present invention, the present invention is further described below with reference to the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification.

[0063] The present invention may also be implemented or applied through other different specific examples, and the details in this specification may also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0064] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly.

[0065] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of such features. Secondly, the technical solutions between the various embodiments can be combined with each other, but this must be based on the fact that ordinary technicians in this field can implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0066] Preferably, the vehicle-mounted collision detection method based on multi-dimensional feature fusion of the present invention is applied to one or more terminals or servers. The terminal is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc.

[0067] The terminal can be a computing device such as a desktop computer, notebook, PDA, cloud server, etc. The terminal can interact with the client through a keyboard, mouse, remote control, touchpad, or voice control device.

[0068] The present invention provides a vehicle-mounted collision detection method, system, platform and storage medium based on multi-dimensional feature fusion.

[0069] like Figure 1 , which is a flow chart of a vehicle-mounted collision detection method based on multi-dimensional feature fusion provided by an embodiment of the present invention.

[0070] In this embodiment, the vehicle-mounted collision detection method based on multi-dimensional feature fusion can be applied to a terminal with a display function or a fixed terminal. The terminal is not limited to a personal computer, a smart phone, a tablet computer, a desktop computer or an all-in-one computer equipped with a camera, etc.

[0071] The vehicle-mounted collision detection method based on multi-dimensional feature fusion can also be applied in a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network, a metropolitan area network, or a local area network. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to embodiments of the present invention can be executed by a server, a terminal, or both.

[0072] For example, for a vehicle-mounted collision detection terminal that needs to perform multi-dimensional feature fusion, the vehicle-mounted collision detection function based on multi-dimensional feature fusion provided by the method of the present invention can be directly integrated on the terminal, or a client for implementing the method of the present invention can be installed. For another example, the method provided by the present invention can also be run on a server or other device in the form of a software development kit (SDK), and an interface for the vehicle-mounted collision detection function based on multi-dimensional feature fusion is provided in the form of the SDK. The terminal or other device can implement the vehicle-mounted collision detection function based on multi-dimensional feature fusion through the provided interface. The present invention is further described below in conjunction with the accompanying drawings.

[0073] like Figure 1 As shown, the present invention provides a vehicle-mounted collision detection method based on multi-dimensional feature fusion, the method comprising the following steps:

[0074] S1. Generate and obtain first data corresponding to the vehicle, and process the first data in segments based on a fixed time window; wherein the first data is three-axis acceleration data of the vehicle;

[0075] S2. Generate and obtain second data corresponding to the first data, and determine and generate third data corresponding to the vehicle collision based on multi-dimensional feature fusion; wherein the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

[0076] The generating and acquiring first data corresponding to the vehicle, and processing the first data in segments in combination with a fixed time window, further includes:

[0077] S11. Collect the vehicle's three-axis acceleration data in real time through the on-board IMU sensor; the sampling frequency shall be no less than 200 Hz;

[0078] S12. Segment processing of the collected acceleration data according to fixed time windows.

[0079] The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0080] S21, calculating and generating fourth data corresponding to the first data, and if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data, executing the next step; wherein the fourth data is the rate of change of acceleration of the three axes;

[0081] S22, calculating and generating ratio data of the Z axis and the XY axis corresponding to the fourth data, and executing the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold;

[0082] S23, calculating and generating the maximum acceleration value of the XY axis, and directly determining a collision when it exceeds the dynamically adjusted acceleration threshold;

[0083] S24, performing wavelet packet decomposition processing on the XY axis acceleration data, calculating the total energy of the 0-6 Hz external frequency band, and proceeding to the next step of determination if it exceeds the threshold;

[0084] S25. Calculate the ratio of the standard deviation to the mean of the high-frequency energy distribution. If the standard deviation exceeds the distribution threshold, it is determined to be a collision event.

[0085] The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0086] S26. Set and generate basic threshold data. For rigidly installed equipment, the basic threshold is set to 500m / s. 3 ;

[0087] S27. For equipment that is not rigidly mounted or is installed on a vehicle with strong energy absorption, the threshold value is lowered; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are lowered proportionally.

[0088] The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0089] S28, detecting the deviation between the Z-axis acceleration and the standard gravity acceleration, calculating the rotation matrix to correct the original three-axis acceleration;

[0090] S29. When the tilt angle exceeds a threshold, a force correction coefficient is introduced for the XY axis acceleration.

[0091] The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes:

[0092] S210, performing wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; wherein a distribution threshold is set to 38, and a collision is determined when the ratio of the energy distribution standard deviation to the mean exceeds this value;

[0093] S211. When a collision is determined to have occurred, automatically upload 90 seconds of video data before and after the collision; and simultaneously upload IMU raw data and vehicle operating status data; the reported data includes vehicle speed, steering angle, gear position, and positioning coordinate information.

[0094] Specifically, in an embodiment of the present invention, a vehicle-mounted collision detection method based on multi-dimensional feature fusion is provided, which achieves high-precision collision detection by combining acceleration, Jerk and wavelet packet energy analysis, and supports dynamic threshold adjustment to adapt to different vehicles and installation conditions.

[0095] The core solution of this invention includes data acquisition and preprocessing: real-time acquisition of three-axis acceleration data (sampling frequency ≥ 200Hz) through the IMU, segmentation processing of the data (fixed time window), and tilt correction (calculation of the rotation matrix based on the Z-axis acceleration and gravity deviation).

[0096] Multiple threshold determination, Jerk threshold determination: calculate the acceleration change rate of adjacent sampling points, if the X / Y axis Jerk exceeds the dynamic threshold (such as 500m / s for rigid installation equipment 3 ), then proceed to the next step of judgment.

[0097] Jerk ratio determination: Calculate the ratio of the Z-axis to the X / Y-axis jerk. If the ratio exceeds the dynamic adjustment threshold (such as 0.8 for a hard-suspension vehicle) or the X / Y-axis jerk exceeds 4 times the basic threshold, the ratio determination is skipped.

[0098] Acceleration threshold judgment: If the maximum acceleration of the X / Y axis exceeds the dynamic threshold (such as 30m / s 2 ), it is directly determined to be a collision.

[0099] Wavelet packet energy analysis: The acceleration data is decomposed by wavelet packet (Daubechies wavelet basis, 4-layer decomposition) to calculate the total energy of the 0-6 Hz external frequency band (the threshold is 1200).

[0100] Calculate the ratio of the standard deviation to the mean of the energy distribution in the high frequency band (the threshold is, for example, 38). If it exceeds this value, it is determined to be a collision.

[0101] Dynamic adjustment and reporting: dynamically adjust the threshold according to vehicle speed and installation angle (such as reducing sensitivity at low speed).

[0102] After a collision is determined, the system automatically uploads 90 seconds of video, IMU data, and vehicle status (speed, steering angle, etc.) before and after the collision.

[0103] In other words, the present invention provides a highly adaptable and sensitive collision detection algorithm that can maintain accurate detection under different installation conditions and different vehicle operators. In addition, when a collision occurs, the device needs to upload key data before and after the accident (such as 90 seconds of video, IMU data, and vehicle status data) to the cloud for tracing and analysis. That is, the present invention addresses the problems of missed detection and false alarms in traditional collision detection, such as Figure 2 As shown in the figure, a collision detection algorithm is proposed which integrates acceleration, Jerk (acceleration change rate) and wavelet packet energy band analysis, which has high adaptability and high sensitivity.

[0104] The analysis of the energy and distribution of the collision acceleration spectrum shows that when a collision occurs in a real environment, the equipment on the vehicle will vibrate under the action of external forces, part of which is caused by impact and part by resonance, such as Figure 3 shown.

[0105] When a collision occurs, the frequency energy distribution generated by operations such as closing a door is more uniform due to the buffering effect, while the frequency energy distribution generated by a rigid collision is more sharp. Here, the vibration spectrum of the collision is expanded, and the energy at different uniformly distributed frequencies is calculated respectively. At the same time, all energy values ​​are used to obtain the normal distribution coefficient of energy. A high coefficient indicates that a collision event has occurred. Figure 4 As shown, data of different frequencies in 5 impacts were obtained by hitting the device with the palm and hitting the chassis with a tool after the device was deployed. The energy distribution of the simulated collision above is sharper and the energy is higher, while the energy distribution of the palm tapping below is more uniform and the energy is lower.

[0106] For deployment and installation, the vehicle-mounted terminal can be installed in a position that fits the vehicle body structure at any angle without affecting the accuracy of the collision detection algorithm; the vehicle-mounted terminal comes with an IMU chip to collect acceleration data, connect to multiple cameras, and obtain real-time data from smart connected vehicles.

[0107] Data acquisition and preprocessing: real-time acquisition of three-axis acceleration (ax, ay, az) via the IMU. The sampling frequency can be set to 200Hz or higher to ensure detection precision.

[0108] The acquired data were segmented into fixed time windows to ensure that signal segments of equal length were used for subsequent analysis.

[0109] Multiple threshold judgment, Jerk threshold: calculate the acceleration difference of adjacent sampling points in the current window and divide it by the sampling period to obtain the three-axis acceleration change rate (Jerk). If extremely high Jerk appears in the horizontal X-axis or vertical Y-axis direction and meets certain threshold requirements, it can be considered that a collision may occur. For different installation positions, the acceleration generated by the force transmitted to the device by the vehicle due to external impact is different. For equipment installed close to the vehicle body, the instantaneous acceleration change rate during a collision is large. The empirical threshold value of the minimum acceleration change rate of the equipment is 500m / s. 3 For devices that are not close to the vehicle body or devices on vehicles with strong energy absorption, the acceleration change rate caused by the collision is small, and a lower threshold is used. The formula for the acceleration change rate in the X-axis direction is as follows: x (i) represents the acceleration change rate for the i-th acceleration sample in the X-axis direction, a x (i) represents the i-th acceleration sample in the X-axis direction, a x (i-1) represents the i-1th acceleration sample in the X-axis direction, and Δt represents sample a x (i) and a x The time difference of (i-1):

[0110]

[0111] Jerk ratio threshold: Calculates the maximum jerk of the three axes within the current window. Compare the maximum jerk of the current Z-axis to the current Y-axis, the current Z-axis to the current X-axis, and the maximum jerk of the previous window's Z-axis to the current Y-axis, and the previous window's Z-axis to the current X-axis. If the ratio exceeds the threshold, the possible collision is determined not to have occurred over a speed bump or pothole. Jerk ratio thresholds vary depending on the installation location and vehicle model. The jerk ratio threshold varies based on the stiffness of the chassis suspension system of different vehicles. A stiff chassis suspension responds quickly and rigidly to road impacts. Setting a low jerk ratio threshold eliminates acceleration noise generated by the chassis during driving. For example, over speed bumps, speed bumps, and potholes can generate large acceleration rates in the Z direction. Based on actual testing, the empirical threshold for the jerk ratio is 0.8. For softer chassis suspension systems, the ratio threshold can be increased. Considering that the transmission of the impact force during a collision may be more distributed in the Z-axis direction, resulting in a low Jerk ratio during a collision, the algorithm skips the Jerk ratio threshold judgment when the Jerk value of the X or Y axis is greater than the empirical value of 4 times the Jerk threshold and directly proceeds to subsequent feature judgment. The formula for the Jerk ratio in the X direction is as follows: ratio is the Jerk ratio value of the X-axis direction to the Z-axis direction within the sampling window, max(|Jerk x (i)|) represents the maximum acceleration change rate in the X-axis direction of the acceleration samples within the sampling window, max(|Jerk z (i)|) represents the maximum acceleration rate of change in the Z-axis direction:

[0112]

[0113] Acceleration threshold: Determine the maximum acceleration of the XY axis within the window. If it exceeds the maximum acceleration threshold, it is considered a collision. If not, continue with the subsequent calculation and threshold judgment of minor collisions and scratches. Since any impact on the vehicle body may produce a momentary high acceleration, a too low acceleration threshold will produce a high number of false alarms. The collision detection here does not rely entirely on a single acceleration threshold judgment. It can be judged by the subsequent acceleration spectrum energy sum or acceleration spectrum energy distribution to determine whether it is a collision. However, if the acceleration generated within the detection window is large enough, it can be directly identified as a collision. The empirical threshold of the maximum acceleration is 30m / s 2 .

[0114] Acceleration spectrum energy summation threshold: Wavelet packet decomposition is performed on the XY-axis acceleration sequence within the window to obtain the decomposed frequency band energy distribution and calculate the summed energy of the frequency bands. If the summed energy of the spectrum outside the 0-6 Hz range exceeds the summed energy threshold, a collision event is identified. The energy in the 0-6 Hz band is filtered out because it often originates from vehicle vibrations caused by uneven roads and aggressive driving behaviors such as sudden acceleration, deceleration, or sharp turns. The characteristics of these low-frequency vibrations generated by everyday driving activities differ significantly from the high-frequency vibrations caused by collisions. The summed energy threshold varies based on the vibration transmission to the equipment after impact on different vehicles in actual deployments. An empirical threshold of 1200 is used here. If the energy is below the minimum threshold, no collision is determined. Otherwise, the subsequent energy distribution calculation is performed to determine a minor collision or scrape scenario. When calculating the acceleration spectrum energy summation, the wavelet packet is based on the Daubechies wavelet object, which facilitates the analysis of sudden signal fluctuations. The decomposition layer is four, decomposing the signal into multiple frequency bands. Each layer further refines the signal, and the period extension method is boundary extension. Through the wavelet packet decomposition tree, the signal is decomposed into different nodes at each layer, and the coefficient value in each node represents the information of the signal within a specific frequency range. The formula is as follows: k represents the energy of the kth frequency band, which is calculated by the sum of the squares of all coefficients in the node of a specific frequency band. E represents the coefficient of the k-th node at the n-th point in the j-th layer of the small cloth bag decomposition. total It is the sum of all frequency band energies, which is obtained by dividing the energy of each frequency band E k Added together. :

[0115]

[0116] Acceleration spectrum energy distribution threshold: Calculate the standard deviation and mean ratio of the XY acceleration spectrum energy distribution outside 0-6Hz in the window. If the energy distribution is greater than the energy distribution threshold, it indicates that the energy of the signal is mainly concentrated in certain frequency bands, and it is judged as a minor collision or scratch event. Otherwise, the energy is relatively evenly distributed among the frequency bands, and it is judged that no collision has occurred in the current window, and the collision detection of the next window will continue. The empirical threshold of the acceleration spectrum energy distribution here is 38, which is the same as the acceleration spectrum energy sum threshold. It varies according to the performance of the vibration transmitted to the equipment after different vehicles are impacted in actual deployment. The formula is as follows, σ E It represents the standard deviation of energy distribution, a statistical indicator that quantifies the uniformity of signal energy distribution and is used to measure the energy E of each frequency band. k Relative to the average energy The degree of discreteness, N represents the total number of frequency bands, E k is the energy of the kth frequency band, This is the energy E of all frequency bandsk The average value of .

[0117]

[0118] Adaptive correction, three-axis acceleration tilt correction: Detect the deviation between the Z-direction acceleration and the regional gravity acceleration constant. A large deviation indicates that the device is installed with a certain tilt. The rotation matrix is ​​calculated based on the difference between the Z-axis acceleration measured by the IMU and the standard gravity, and then the original three-axis acceleration is corrected accordingly to reduce the deviation of the horizontal and vertical acceleration measurements. For devices that are tilted too much and exceed the threshold and are deployed leaning, when a force is applied in the Z-axis direction, part of the force will be distributed to the XY axis. A force correction coefficient is introduced according to the leaning angle to adjust the final acceleration value of XY and reduce misjudgment. The correction coefficient formula for the acceleration direction is as follows, where Cx is the X-axis gravity correction coefficient, Ax is the original X-axis acceleration, Az is the original Z-axis acceleration, θ is the sensor tilt angle, and g is the gravity acceleration:

[0119]

[0120] Low-speed driving threshold adjustment: If the current vehicle speed is less than the minimum speed threshold, the acceleration noise caused by the external environment such as road bumps is small, lowering the overall detection threshold and increasing sensitivity.

[0121] Results and data reporting: When the algorithm comprehensively determines a collision has occurred, an event report is triggered. 90 seconds of video footage, IMU raw data, and vehicle operating data (speed, cornering angle, gear position, positioning coordinates, etc.) before and after the collision are simultaneously uploaded to the server. The server uses the device communication protocol to adjust the detection threshold in real time based on the reported collision event data.

[0122] The example of collision detection for electric scooters is as follows: on a regulated autonomous driving application test operation vehicle, the equipment is deployed on a dedicated equipment rack in the modified trunk. The equipment rack is rigidly connected to the modified trunk and the vehicle body, and the vehicle chassis suspension is relatively soft. When the accident occurred, there were no passengers on the vehicle. The vehicle drove out of the highway gate toll station and tried to merge into the main road at a low speed at a T-junction. An electric scooter was rapidly approaching the vehicle outside the blind spot of the vehicle on the left side of the intersection. Because the electric scooter was traveling too fast, the autonomous driving system and the safety officer identified the electric scooter and collided with the electric scooter when they tried to brake. The equipment detected the collision accident and reported it to the platform. The collision detection threshold of the device is, the minimum Jerk threshold is 500m / s 3 , the minimum Jerk ratio threshold is 0.85, and the maximum acceleration threshold is 30m / s 2 The maximum acceleration spectrum energy sum threshold is 1200, the maximum acceleration spectrum energy distribution threshold is 38, the device sampling rate is 200Hz, and the device installation tilt angle is 6 degrees. When the collision occurs, the vehicle speed is less than 5m / s, and the maximum acceleration threshold is dynamically adjusted to 15m / s.2 The minimum Jerk ratio threshold was dynamically adjusted to 0.425. The final collision detection algorithm determined a collision based on the acceleration spectrum energy distribution values, with the energy spectrum distribution coefficient in the X direction being 42.3 and the energy distribution coefficient in the Y direction being 38.6. Below is the one-second acceleration sample data generated by the sampling window during this collision accident, containing 200 corrected acceleration values ​​(ax, ay, az) for the X, Y, and Z axes, respectively:

[0123] [{-0.15,-9.12,7.84},{-0.27,-9.45,3.95},{0.46,-9.54,17.62},{1.19,-8.95,5.48},{0.81,-9.63,-2.42},{0.21,-8.48,11.49},{-0.72,-8.77,9.3},{-0.6,-8.17,2.35},{-0.16,-8.97,12.84},{-0.05,-9.36,10.06},{0.34,-9.04,-0.7},{2.2,-9.3,12.9},{1.72,-9.14,14.46},{0.47,-8.66,1.99},{0.59,-8.73,12.34},{0.14,-9.38,13.61},{-0.52,-8.16,-0.18},{-0.06,-8.82,9.18},{0.38,-9.32,16.68},{1.04,-8.46,6.61},{1.48,-9.04,9.89},{1.23,-9.23,10.83},{0.37,-8.97,0.94},{0.4,-9.3,6.22},{0.8,-9.35,13.47},{0.74,-8.49,5.04},{0.41,-9.34,6.09},{0.99,-8.89,11.73},{1.48,-8.51,4.63},{1.46,-9.01,4.41},{0.87,-9.76,11.78},{0.14,-9.28,6.01},{-0.43,-9.91,0.96},{0.39,-9.52,9.77},{0.34,-9.75,13.08},{-0.27,-9.16,8.1},{0.45,-9.34,8.96},{1.25,-10.28,13.04},{1.58,-9.49,5.99},{1.17,-10.02,6.68},{0.17,-10.23,13.63},{0.16,-9.8,5.99},{0.2,-10.19,4.88},{0.69,-10.41,9.97},{0.89,-10.18,9.21},{0.17,-10.41,12.23},{-0.03,-10.25,10.44},{0.12,-10.57,8.01},{0.17,-10.76,8.48},{0.7,-10.4,10.43},{0.42,-10.23,13.5},{0.1,-10.01,6.99},{0.71,-9.52,7.96},{0.77,-8.98,11.72},{0.85,-8.17,6.49},{0.31,-7.86,13.19},{0.07,-7.16,10.43},{0.31,-6.39,4.48},{0.02,-6.11,13.09},{0.21,-5.24,9.85},{0.35,-4.54,6.89},{0.61,-3.96,13.14},{0.79,-3.28,7.99},{0.33,-2.49,7.43},{0.81,-2.19,11.86},{0.58,-2.44,8.92},{0.4,-1.53,8.61},{1.45,0.34,11.95},{1.73,-0.57,10.96},{-0.63,0.44,7.85},{0.31,0.24,11.56},{0.28,0.76,14.4},{1.13,-1.26,6.34},{-1.74,-1.87,8.36},{-5.2,-3.94,13.72},{-4.22,2.46,11.14},{9.42,-1.25,2.09},{9.3,-5.34,14.68},{-3.52,-3.82,32.51},{-4.3,-4.28,5.82},{1.66,-2.22,18.68},{10.16,-0.84,4.79},{5.77,-0.08,2.42},{-2.83,-1.05,35.41},{0.28,-1.42,7.79},{6.17,-0.29,15.58},{5.56,-0.08,14.84},{1.98,0.72,4.24},{1.07,3.36,14.26},{3.59,4.47,11.31},{2.09,6,0.41},{0.79,4.61,13.78},{0.79,5.25,11.45},{1.84,6.97,-2.11},{2.83,6.15,8.23},{1.8,6.92,15.17},{1.2,8.93,-0.05},{2.25,8.95,12.64},{2.94,7.85,17.78},{1.76,7.99,0.31},{1.84,7.04,13.9},{2.33,7.22,17.58},{2.43,8.4,7.11},{2.96,7.86,13.9},{2.91,7.22,14.5},{2.74,6.87,5.15},{3.1,6.37,12.61},{3.63,7.17,9.57},{3.3,7.07,3.68},{1.79,6.98,13.97},{1.69,6.79,7.89},{2.44,6.28,0.48},{2.5,4.72,15.06},{1.82,4.7,3.55},{1.24,4.45,1.81},{1.47,3.74,15.45},{1.46,4.03,5.34},{0.94,3.25,5.84},{0.34,2.44,13.76},{0.24,2.34,5.88},{-0.2,1.62,6.51},{-0.54,0.98,12.22},{-1.05,0.94,7.07},{-1.01,0.49,7.34},{-0.74,-0.12,13.12},{-1.39,-0.17,7.38},{-2.39,-1.01,7},{-2.47,-1.13,11.7},{-2.76,-1.5,7.62},{-3.31,-1.82,6.34},{-3.55,-2.18,11.23},{-3.67,-2.43,8.34},{-4.01,-2.79,5.3},{-3.99,-2.89,7.54},{-4.38,-3.34,7.14},{-4.49,-3.57,7.52},{-4.73,-3.89,9.32},{-5.02,-4.08,9.57},{-5.29,-4.25,8.37},{-4.98,-4.32,8.6},{-4.7,-4.58,9.67},{-4.81,-4.5,8.43},{-4.75,-4.62,8.82},{-4.72,-4.86,10.9},{-4.72,-4.91,7.86},{-4.64,-4.85,7.03},{-4.17,-4.78,10.02},{-3.74,-4.64,9.2},{-3.74,-4.42,7.68},{-3.42,-4.32,9.85},{-2.91,-4.29,10.47},{-2.56,-3.85,9.25},{-2.1,-3.72,9.64},{-2.02,-3.5,10.55},{-1.58,-3.3,8.4},{-1.04,-3.31,9.89},{-0.46,-3.01,10.07},{0.14,-2.59,8.9},{0.51,-2.29,11.54},{0.74,-1.98,9.81},{1.13,-1.54,9.41},{1.45,-1.32,13.24},{1.92,-0.96,9.02},{2.34,-0.91,10},{2.79,-0.54,12.81},{3.09,-0.3,9.33},{3.38,0.04,11.42},{3.9,0.1,12.45},{4.22,0.54,9.17},{4.38,0.55,12.02},{4.64,0.93,11.9},{4.72,1.21,10.61},{4.93,1.32,12.7},{4.8 9,1.54,12.25},{5.19,1.7,10.21},{5.26,1.81,12.26},{5.34,1.99,11.48},{5.41,2.13,11.08},{5.59,2.19,12.34},{5.56,2.29,11.27},{5.36,2.41,10.8},{5.35,2.3,12.97},{5. 37,2.41,10.87},{5.18,2.43,11.74},{4.86,2.26,12.56},{4.76,2.21,10.28},{4.54,2.06,11.95},{4.43,1.92,11.86},{4.25,1.84,10.12},{3.99,1.65,12.25},{3.75,1.61,11.13} ,{3.53,1.51,10.31},{3.26,1.21,12.4},{3.03,1.01,10.51},{2.72,0.84,10.34},{2.44,0.63,11.79},{2.28,0.5,9.89},{1.98,0.32,10.43},{1.79,0.18,11.14},{1.52,0.1,9.29}].

[0124] According to the collision algorithm, the maximum Jerk on the X axis within the sampling window is 2728m / s 3 , the maximum jerk on the Y axis is 1280m / s 3 , are greater than the minimum Jerk threshold of 500m / s 3 , it is judged that a collision is likely to occur, and subsequent collision detection is performed.

[0125] Z-axis Jerk: 6598m / s 3 The Jerk ratio of the X axis is 0.41, and the Jerk ratio of the Y axis is 0.19, both of which are less than the dynamically adjusted Jerk ratio threshold of 0.425. However, the maximum Jerk of the X axis is 2728m / s. 3 Greater than 4 times the Jerk threshold of 2000m / s 3 , the algorithm continues with subsequent collision detection.

[0126] In the maximum acceleration threshold judgment, the maximum value of the X axis is 10.16m / s 2 , the maximum value of the Y axis is 10.76m / s 2 , are all lower than the maximum acceleration threshold of 15m / s after dynamic adjustment 2 ,The algorithm continues with subsequent collision detection.

[0127] The wavelet packet decomposition algorithm calculates the acceleration spectrum energy in the X-axis and Y-axis directions as shown in the table below. The total energy on the X-axis is 456, and the total energy on the Y-axis is 306. Both values ​​do not exceed the maximum acceleration spectrum energy sum threshold of 1200. The algorithm then proceeds with subsequent collision calculations.

[0128] Table 1 Acceleration spectrum energy table of the embodiment

[0129]

[0130]

[0131] The acceleration spectrum energy distribution algorithm calculated that the X-axis energy distribution was 42.38 and the Y-axis energy distribution was 38.63, both exceeding the maximum acceleration spectrum energy distribution threshold of 38. The algorithm detected the collision, generated collision accident data, and uploaded the preceding and following 90 seconds of video data and accelerometer data to the monitoring platform. The event is visible on the platform.

[0132] To achieve the above object, the present invention also provides a vehicle-mounted collision detection system based on multi-dimensional feature fusion, wherein the system is used to implement the vehicle-mounted collision detection method based on multi-dimensional feature fusion; Figure 10 As shown, the system specifically includes:

[0133] A first data generating unit is configured to generate and obtain first data corresponding to the vehicle, and process the first data in segments in combination with a fixed time window; wherein the first data is three-axis acceleration data of the vehicle;

[0134] The second data generation unit is used to generate and obtain second data corresponding to the first data, and based on multi-dimensional feature fusion, determine and generate third data corresponding to the vehicle collision; wherein, the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

[0135] The first data generating unit further includes:

[0136] The first processing module is used to collect the vehicle's three-axis acceleration data in real time through the vehicle-mounted IMU sensor; wherein the sampling frequency is not less than 200 Hz;

[0137] The second processing module is used to process the collected acceleration data in segments according to fixed time windows;

[0138] And / or, the second data generating unit further includes:

[0139] a first generating module, configured to calculate and generate fourth data corresponding to the first data, and execute the next step if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data; wherein the fourth data is a rate of change of acceleration of the three axes;

[0140] a second generating module, configured to calculate and generate ratio data of the Z axis and the XY axis corresponding to the fourth data, and execute the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold;

[0141] The third generation module is used to calculate and generate the maximum acceleration value of the XY axis, and directly determine it as a collision when it exceeds the dynamically adjusted acceleration threshold;

[0142] The third processing module is used to perform wavelet packet decomposition on the XY axis acceleration data and calculate the total energy of the 0-6Hz external frequency band. If it exceeds the threshold, it will proceed to the next step of judgment;

[0143] The first determination module is used to calculate the standard deviation and mean ratio of the high-frequency energy distribution, and determine it as a collision event if it exceeds the distribution threshold.

[0144] The second data generating unit further includes:

[0145] The fourth generation module is used to set and generate basic threshold data, wherein the basic threshold for rigidly installed equipment is set to 500m / s 3 ;

[0146] The fifth generation module is used to lower the threshold value for devices that are not rigidly mounted or have strong energy absorption on the vehicle; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are proportionally lowered;

[0147] The fourth processing module is used to detect the deviation between the Z-axis acceleration and the standard gravity acceleration, calculate the rotation matrix and correct the original three-axis acceleration;

[0148] a fifth processing module, configured to introduce a force correction coefficient into the XY axis acceleration when the tilt angle exceeds a threshold;

[0149] The sixth processing module is used to perform wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; wherein the distribution threshold is set to 38, and a collision is determined when the ratio of the energy distribution standard deviation to the mean exceeds this value;

[0150] The seventh processing module is used to automatically upload 90 seconds of video data before and after the collision when a collision is determined to have occurred; and simultaneously upload the IMU raw data and vehicle operating status data; among which the reported data includes vehicle speed, steering angle, gear position and positioning coordinate information.

[0151] In the system solution embodiment of the present invention, the specific details of the method steps involved in the vehicle-mounted collision detection based on multi-dimensional feature fusion have been explained above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, which will not be repeated here.

[0152] To achieve the above objectives, the present invention also provides a vehicle-mounted collision detection platform based on multi-dimensional feature fusion, such as Figure 11 As shown, it includes a processor, a memory, and a vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion; wherein, the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is executed by the processor, and the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is stored in the memory, and the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion implements the vehicle-mounted collision detection method steps based on multi-dimensional feature fusion. For example:

[0153] S1. Generate and obtain first data corresponding to the vehicle, and process the first data in segments based on a fixed time window; wherein the first data is three-axis acceleration data of the vehicle;

[0154] S2. Generate and obtain second data corresponding to the first data, and determine and generate third data corresponding to the vehicle collision based on multi-dimensional feature fusion; wherein the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

[0155] The specific details of the steps have been explained above and will not be repeated here.

[0156] In an embodiment of the present invention, the built-in processor of the vehicle-mounted collision detection platform based on multi-dimensional feature fusion can be composed of an integrated circuit, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor utilizes various interfaces and circuits to connect various components, and executes or runs programs or units stored in the memory, as well as calls data stored in the memory, to perform various functions of the vehicle-mounted collision detection based on multi-dimensional feature fusion and process data.

[0157] The memory is used to store program codes and various data. It is installed in the vehicle-mounted collision detection platform based on multi-dimensional feature fusion, and realizes high-speed and automatic access to programs or data during operation.

[0158] The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electronically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0159] The present invention generates and obtains first data corresponding to a vehicle through a method, and processes the first data in segments in combination with a fixed time window; wherein the first data is the vehicle's three-axis acceleration data; generates and obtains second data corresponding to the first data, and determines and generates third data corresponding to a vehicle collision based on multi-dimensional feature fusion; wherein the second data is a preset dynamic threshold; the third data is vehicle collision event detection data, as well as a system and platform corresponding to the method, which achieves high-precision collision determination by fusing acceleration, Jerk, and wavelet packet energy analysis. The technical solution includes data acquisition, multiple threshold determination (Jerk, ratio, acceleration, frequency domain energy), and dynamic adjustment, significantly reducing false alarm and missed detection rates, and is suitable for safety monitoring of intelligent connected vehicles.

[0160] In other words, the solution of the present invention adapts to different installation conditions and vehicle types through dynamic threshold adjustment and tilt correction. Combining Jerk ratio and wavelet packet energy analysis, it effectively filters out interference such as road bumps. Frequency domain analysis can capture minor collisions, reducing the missed detection rate. Furthermore, the solution of the present invention has low algorithm complexity, making it suitable for deployment on vehicle terminals.

[0161] In other words, the solution of this invention is highly adaptable: it doesn't rely on a single acceleration threshold, but instead integrates multiple criteria, including lateral, longitudinal, and multi-band energy. It can adapt to various installation angles, reducing IMU sensor data discrepancies caused by installation and deployment variations, thereby reducing missed detections and false alarms in collision events, facilitating real-time capture and monitoring of collisions across different automakers and operators. Furthermore, for unique installation and deployment scenarios, the sensitivity can be dynamically adjusted through online threshold correction, better adapting to different vehicles and operating environments.

[0162] High sensitivity and low false alarm rate: Jerk (acceleration rate of change) and frequency domain analysis can capture rapid, instantaneous changes, significantly reducing missed detections of low-intensity impacts and false alarms caused by chassis-directional impacts. Specifically, the ratio of Jerk on the Z-axis to the XY-axis filters out false alarms caused by complex road conditions and chassis-directional forces. Wavelet packet decomposition of the acceleration signal calculates the energy distribution and total energy of each frequency band coefficient within different frequency bands, identifying minor collisions and scrapes while reducing missed detection rates. This is further reduced by combining the low-speed threshold reduction correction.

[0163] Facilitates vehicle data management: When a collision is detected, the system automatically reports IMU data, video data, and vehicle status, enabling accident scene playback and liability determination. The server can also modify the device's detection threshold online based on the collision detection results to accommodate more complex scenarios.

[0164] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A vehicle collision detection method based on multi-dimensional feature fusion, characterized in that: The method comprises: Generate and obtain first data corresponding to the vehicle, and process the first data in segments based on a fixed time window; wherein the first data is three-axis acceleration data of the vehicle; Generate and obtain second data corresponding to the first data, and based on multi-dimensional feature fusion, determine and generate third data corresponding to the vehicle collision; wherein, the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

2. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to claim 1, characterized in that: The generating and acquiring first data corresponding to the vehicle, and processing the first data in segments in combination with a fixed time window, further includes: The vehicle's three-axis acceleration data is collected in real time through the on-board IMU sensor, with a sampling frequency of no less than 200Hz; The collected acceleration data is segmented and processed according to fixed time windows.

3. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to claim 1, characterized in that: The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes: Calculating and generating fourth data corresponding to the first data, and executing the next step if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data; wherein the fourth data is the rate of change of acceleration of the three axes; Calculate and generate ratio data of the Z axis and the XY axis corresponding to the fourth data, and execute the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold; Calculate and generate the maximum acceleration value of the XY axis. When it exceeds the dynamically adjusted acceleration threshold, it is directly determined as a collision; Perform wavelet packet decomposition on the XY axis acceleration data and calculate the total energy of the 0-6Hz external frequency band. If it exceeds the threshold, proceed to the next step of judgment. The standard deviation to mean ratio of the high-frequency energy distribution is calculated, and if it exceeds the distribution threshold, it is determined to be a collision event.

4. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to claim 1 or 3, characterized in that: The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes: Set the generated basic threshold data. For rigidly mounted equipment, the basic threshold is set to 500m / s. 3 ; For equipment that is not rigidly installed or on vehicles with strong energy absorption, the threshold processing is lowered; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are reduced proportionally.

5. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to claim 4, characterized in that: The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes: Detect the deviation between the Z-axis acceleration and the standard gravity acceleration, calculate the rotation matrix and correct the original three-axis acceleration; When the tilt angle exceeds the threshold, a force correction coefficient is introduced for the XY axis acceleration.

6. The vehicle-mounted collision detection method based on multi-dimensional feature fusion according to claim 4, characterized in that: The generating and acquiring second data corresponding to the first data, and determining and generating third data corresponding to the vehicle collision based on multi-dimensional feature fusion, further includes: Perform wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; set the distribution threshold to 38, and determine a collision when the ratio of the energy distribution standard deviation to the mean exceeds this value; When a collision is detected, the system automatically uploads 90 seconds of video data before and after the collision, and simultaneously uploads IMU raw data and vehicle operating status data. The reported data includes vehicle speed, steering angle, gear position, and positioning coordinate information.

7. A vehicle-mounted collision detection system based on multi-dimensional feature fusion, characterized in that: The system is applied to a vehicle-mounted collision detection method based on multi-dimensional feature fusion as described in any one of claims 1 to 6; the system comprises: A first data generating unit is configured to generate and obtain first data corresponding to the vehicle, and process the first data in segments in combination with a fixed time window; wherein the first data is three-axis acceleration data of the vehicle; The second data generation unit is used to generate and obtain second data corresponding to the first data, and based on multi-dimensional feature fusion, determine and generate third data corresponding to the vehicle collision; wherein, the second data is a preset dynamic threshold; and the third data is vehicle collision event detection data.

8. The vehicle-mounted collision detection system based on multi-dimensional feature fusion according to claim 7, characterized in that: The first data generating unit further includes: The first processing module is used to collect the vehicle's three-axis acceleration data in real time through the vehicle-mounted IMU sensor; wherein the sampling frequency is not less than 200 Hz; The second processing module is used to process the collected acceleration data in segments according to fixed time windows; And / or, the second data generating unit further includes: a first generating module, configured to calculate and generate fourth data corresponding to the first data, and execute the next step if the fourth data corresponding to the X-axis or the Y-axis exceeds the second data; wherein the fourth data is a rate of change of acceleration of the three axes; a second generating module, configured to calculate and generate ratio data of the Z axis and the XY axis corresponding to the fourth data, and execute the next step when the ratio exceeds a dynamically adjusted ratio threshold or the fourth data of the XY axis exceeds 4 times a basic threshold; The third generation module is used to calculate and generate the maximum acceleration value of the XY axis, and directly determine it as a collision when it exceeds the dynamically adjusted acceleration threshold; The third processing module is used to perform wavelet packet decomposition on the XY axis acceleration data and calculate the total energy of the 0-6Hz external frequency band. If it exceeds the threshold, it will proceed to the next step of judgment; The first determination module is used to calculate the standard deviation and mean ratio of the high-frequency energy distribution, and determine it as a collision event if it exceeds the distribution threshold.

9. The vehicle-mounted collision detection system based on multi-dimensional feature fusion according to claim 7 or 8, characterized in that: The second data generating unit further includes: The fourth generation module is used to set and generate basic threshold data, wherein the basic threshold for rigidly installed equipment is set to 500m / s 3 ; The fifth generation module is used to lower the threshold value for devices that are not rigidly mounted or have strong energy absorption on the vehicle; when the vehicle speed is lower than the preset minimum speed, all judgment thresholds are proportionally lowered; The fourth processing module is used to detect the deviation between the Z-axis acceleration and the standard gravity acceleration, calculate the rotation matrix and correct the original three-axis acceleration; a fifth processing module, configured to introduce a force correction coefficient into the XY axis acceleration when the tilt angle exceeds a threshold; The sixth processing module is used to perform wavelet packet energy distribution statistical analysis on typical collision events and non-collision events; wherein the distribution threshold is set to 38, and a collision is determined when the ratio of the energy distribution standard deviation to the mean exceeds this value; The seventh processing module is used to automatically upload 90 seconds of video data before and after the collision when a collision is determined to have occurred; and simultaneously upload the IMU raw data and vehicle operating status data; among which the reported data includes vehicle speed, steering angle, gear position and positioning coordinate information.

10. A vehicle-mounted collision detection platform based on multi-dimensional feature fusion, characterized in that: The invention comprises a processor, a memory and a vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion; wherein, the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is executed by the processor, the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion is stored in the memory, and the vehicle-mounted collision detection platform control program based on multi-dimensional feature fusion implements the vehicle-mounted collision detection method based on multi-dimensional feature fusion as described in any one of claims 1 to 6.

Images