Collision detection method and device based on multi-sensor data fusion and storage medium

Abstract

The invention relates to a collision detection method and device based on multi-sensor data fusion, and the method comprises the steps: collecting the dynamic information of a vehicle in real time through a plurality of sensors disposed on the vehicle; preprocessing the dynamic information collected by the plurality of sensors, and storing the preprocessed data; automatically adjusting a collision detection strategy and a threshold value of the deep learning model according to the real-time driving environment and the vehicle state of the vehicle; performing comprehensive analysis and collision detection; once a collision is detected, video records, sensor data and related parameters before and after the collision are automatically stored. By fusing various sensor data, the limitation of dependence on a single data source is reduced, and the accuracy of collision prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of collision detection, and in particular to a collision detection method, device and storage medium based on multi-sensor data fusion. Background Art

[0002] As the demand for driving safety increases, the functionality of dashcams is constantly evolving. Traditional dashcams rely primarily on video recording but lack effective collision detection and response mechanisms. Existing collision detection technologies primarily rely on single sensors, such as accelerometers, which often fail to fully reflect the vehicle's state under complex road conditions, resulting in insufficient collision detection accuracy. Summary of the Invention

[0003] The present invention provides a collision detection method, device and storage medium based on multi-sensor data fusion, aiming to solve at least one of the technical problems existing in the prior art.

[0004] The technical solution of the present invention is a collision detection method based on multi-sensor data fusion, which is applied to a collision detection device based on multi-sensor data fusion. The collision detection device based on multi-sensor data fusion includes a data acquisition unit, which includes at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing vehicle geographic location information and speed data, and a steering angle sensor for detecting the steering angle of the steering wheel; a data processing unit, which is based on a SOC (System on Chip) platform and includes at least a central processing unit (CPU) and a neural network processor (NPU); an alarm and recording unit, which performs collision detection and then generates alarm information to the board end and mobile end APP at the same time, and saves video data to a local storage card or cloud space. The collision detection method based on multi-sensor data fusion includes the following steps:

[0005] S100, collects vehicle dynamic information in real time through multiple sensors installed on the vehicle;

[0006] S200, preprocessing the dynamic information collected by multiple sensors and saving the preprocessed data;

[0007] S300, automatically adjusting the collision detection strategy and threshold of the deep learning model according to the real-time driving environment and vehicle status of the vehicle;

[0008] S400, comprehensive analysis and collision detection: Based on the parameters and threshold results adjusted in real time, the pre-processed sensor data is comprehensively analyzed to determine the collision detection conditions. When it is detected that the acceleration, angular velocity, and steering angle meet the collision threshold, the system determines that a collision has occurred;

[0009] S500: Once a collision is detected, automatically save the video records, sensor data and related parameters before and after the collision.

[0010] Furthermore, in step S100,

[0011] The multiple sensors include at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting the steering angle of the steering wheel.

[0012] Furthermore, the step S200 includes:

[0013] S210, synchronously aligning the timestamps of the sensor data to ensure that the data are consistent in time;

[0014] S220, using a Kalman filter and Gaussian smoothing to denoise multiple sets of sensor data and filter out abnormal data points;

[0015] S230. Perform calculations based on the filtered data, including vehicle speed, acceleration, angle between initial acceleration and acceleration during driving, position change, and time interval, and save the calculation results in chronological order.

[0016] Furthermore, the step S220 includes:

[0017] S221, first use the Kalman filter to recursively estimate the state from the initial moment to the final moment;

[0018] S222: After the forward filtering is completed, the obtained Kalman filtering result is used to perform backward smoothing;

[0019] S223. Recursively calculate from the last moment forward, adjust the state estimate at each moment, and update the state at the previous moment in combination with the state information at the subsequent moment to further reduce the error.

[0020] Furthermore, in step S230,

[0021] Using the location information provided by the GPS device, the driving speed is calculated in real time. Based on the Haversine formula, the surface distance d between two points is calculated. Assume that two consecutive location points P1 and P2 are obtained from the GPS device. The two location points include longitude and latitude coordinates and timestamps P1 (longitude λ1, latitude φ1, time t1) and P2 (longitude λ2, latitude φ2, time t2). Substitute into:

[0022]

[0023] Where φ and λ are in radians, R is the radius of the Earth, d is the distance between two points, and velocity v is the rate of change of distance over time. The formula for calculating velocity v is:

[0024]

[0025] Based on the data provided by the GPS device, the vehicle's acceleration value per second is calculated in real time. Severe negative acceleration indicates that the vehicle has braked suddenly or collided. Three consecutive positions P1, P2, and P3 are obtained from the GPS device, corresponding to time points t1, t2, and t3, respectively. First, the velocity v formula is applied to calculate the velocity v1 from P1 to P2 and the velocity v2 from P2 to P3. Then, the acceleration a is calculated:

[0026]

[0027] The angle α between the initial acceleration and the acceleration during driving is:

[0028]

[0029] in, is the initial acceleration value when the vehicle starts, is the acceleration value of the vehicle during driving;

[0030] In step S230, the position changes of the vehicle are monitored in real time through the GPS device data. If the vehicle suddenly stops during driving and the position hardly changes, the speed and acceleration of the vehicle are combined to assist in determining whether a collision has occurred. Collisions usually occur in a very short period of time. By monitoring the time interval of speed changes, it is assisted in determining whether a collision has occurred.

[0031] Furthermore, the angle α between the initial acceleration and the acceleration during driving is in the range of 40-55 degrees.

[0032] Furthermore, the step S300 includes:

[0033] S310, using an on-board camera image to identify the driving environment and weather conditions, and identifying the vehicle's driving state based on the vehicle's speed, acceleration, GPS coordinates, and steering angle, where the driving environment includes at least the GPS location, and the vehicle's driving state includes at least normal driving, sudden acceleration, sudden deceleration, and sharp turn;

[0034] S320, analyzing the current driving environment, weather conditions, and vehicle driving status through a deep learning model, and dynamically adjusting collision detection thresholds and parameters;

[0035] S330. Update parameter settings based on real-time data to optimize collision detection performance, and adjust thresholds based on detection results to reduce false positives and false negatives.

[0036] Furthermore, in step S320,

[0037] The deep learning model includes a lightweight neural network model MobileNet and a long short-term memory network LSTM;

[0038] The neural network model MobileNet is used to identify driving environment and weather conditions from images from the vehicle camera, and the long short-term memory network LSTM is used to analyze the vehicle driving status.

[0039] Furthermore, the present invention also proposes a collision detection device based on multi-sensor data fusion, which is used to execute the collision detection method based on multi-sensor data fusion. The collision detection device based on multi-sensor data fusion includes:

[0040] a data acquisition unit, the data acquisition unit comprising at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting a steering angle of a steering wheel;

[0041] A data processing unit, which is based on a SOC (System on Chip) platform and includes at least a central processing unit (CPU) and a neural network processor (NPU);

[0042] The alarm and recording unit and the data processing unit perform collision detection, and then generate alarm information to the board and mobile APP at the same time, and save the video data to the local storage card or cloud space.

[0043] Furthermore, the present invention also proposes a computer-readable storage medium having program instructions stored thereon, and when the program instructions are executed by a processor, the collision detection method based on multi-sensor data fusion is implemented.

[0044] The beneficial effects of the present invention are:

[0045] The multi-sensor data fusion-based collision detection method reduces the limitations of relying on a single data source by fusing multiple sensor data, thereby improving the accuracy of collision prediction. A real-time and rapid processing and response mechanism ensures timely warnings and preservation of video data at critical moments. The method enhances the functionality of the dashcam, enabling not only recording but also proactive prevention and analysis of collision events. The method can operate effectively in all weather and lighting conditions, offering wide adaptability and ensuring driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is the overall flow chart of the collision detection method based on multi-sensor data fusion.

[0047] Figure 2 The present invention is a flowchart for preprocessing dynamic information collected by multiple sensors and saving the preprocessed data in a collision detection method based on multi-sensor data fusion.

[0048] Figure 3 This is a flowchart for automatically adjusting the collision detection strategy and threshold of the deep learning model according to the vehicle's real-time driving environment and vehicle status in the collision detection method based on multi-sensor data fusion.

[0049] Figure 4 Schematic diagram of a collision detection device based on multi-sensor data fusion. DETAILED DESCRIPTION

[0050] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless there is a conflict.

[0051] It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature or indirectly fixed or connected to the other feature. Furthermore, terms such as "upper," "lower," "left," "right," "top," and "bottom" used in this disclosure are intended solely to describe the relative positions of the components of the disclosure as shown in the accompanying drawings.

[0052] In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art. The terms used in this specification are only for describing specific embodiments and are not intended to limit the present invention. The term "and / or" as used herein includes any combination of one or more of the related listed items.

[0053] It should be understood that although the terms first, second, third, etc. may be used to describe various elements in the present disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element, and similarly, a second element may also be referred to as a first element without departing from the scope of the present disclosure.

[0054] Reference Figures 1 to 4 In some embodiments, the technical solution of the present invention is a collision detection method based on multi-sensor data fusion, and the collision detection method based on multi-sensor data fusion is applied to a collision detection device based on multi-sensor data fusion, and the collision detection device based on multi-sensor data fusion includes a data acquisition unit, and the data acquisition unit includes at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing vehicle geographic location information and speed data, and a steering angle sensor for detecting the steering angle of the steering wheel; a data processing unit, and the data processing unit is based on a SOC (System on Chip, system on chip) platform, and includes at least a central processing unit CPU and a neural network processor NPU; an alarm and recording unit, the data processing unit performs collision detection, and then generates alarm information to the board end and the mobile end APP at the same time, and saves the video data to the local storage card or cloud space, refer to Figure 1 , the collision detection method based on multi-sensor data fusion includes the following steps:

[0055] S100, collects vehicle dynamic information in real time through multiple sensors installed on the vehicle;

[0056] S200, preprocessing the dynamic information collected by multiple sensors and saving the preprocessed data;

[0057] S300, automatically adjusting the collision detection strategy and threshold of the deep learning model according to the real-time driving environment and vehicle status of the vehicle;

[0058] S400, comprehensive analysis and collision detection: Based on the parameters and threshold results adjusted in real time, the pre-processed sensor data is comprehensively analyzed to determine the collision detection conditions. When it is detected that the acceleration, angular velocity, and steering angle meet the collision threshold, the system determines that a collision has occurred;

[0059] S500: Once a collision is detected, automatically save the video records, sensor data and related parameters before and after the collision.

[0060] The beneficial effects of the present invention are:

[0061] The multi-sensor data fusion-based collision detection method improves collision prediction accuracy by fusing data from multiple sensors, reducing the limitations of relying on a single data source. A real-time, rapid processing and response mechanism ensures timely warnings and video data storage at critical moments. This enhances the functionality of dashcams, enabling not only recording but also proactive collision prevention and analysis. The method operates effectively in all weather and lighting conditions, offering wide adaptability and ensuring driving safety. The present invention relates to dashcams and vehicle collision detection technology, specifically a method that utilizes multi-sensor fusion technology to improve the accuracy and response speed of vehicle collision detection in diverse environments and conditions. This method integrates a six-axis sensor (a three-axis accelerometer and a three-axis gyroscope), GPS, and a steering angle sensor, combined with advanced data processing and machine learning algorithms, to accurately detect and record vehicle collisions. This method provides a simple and convenient method that leverages multi-sensor data fusion technology and machine learning algorithms to accurately detect vehicle collisions and simultaneously store emergency video, overcoming the problem of existing collision detection methods, where excessive invalid detections lead to large storage space requirements for emergency video. The present invention is applicable to any emergency situation encountered while driving that requires video storage.

[0062] Further, refer to Figure 1 In the step S100,

[0063] The multiple sensors include at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting the steering angle of the steering wheel.

[0064] Further, refer to Figure 2 , the step S200 includes:

[0065] S210, synchronously aligning the timestamps of the sensor data to ensure that the data are consistent in time;

[0066] S220, using a Kalman filter and Gaussian smoothing to denoise multiple sets of sensor data and filter out abnormal data points;

[0067] S230. Perform calculations based on the filtered data, including vehicle speed, acceleration, angle between initial acceleration and acceleration during driving, position change, and time interval, and save the calculation results in chronological order.

[0068] Specifically, in step S220, for filtering, a Kalman filter is first used to recursively estimate the state from the initial moment to the final moment. After the forward filtering is completed, the obtained Kalman filter result is used for backward smoothing, and recursive calculation is performed from the final moment forward to adjust the state estimate at each moment. This part will update the state of the previous moment in combination with the state information of the subsequent moment to further reduce the error. In order to better understand the filtering process, the following is explained with reference to the specific formula:

[0069] (1) State prediction

[0070]

[0071] (2) Result update

[0072]

[0073] (3) Gaussian smoothing

[0074]

[0075] Among them, x k-1 and x k The posterior state estimates at time k-1 and time k are respectively one of the filtering results, that is, the updated result; x k represents the prior state estimate at time k, that is, the state at time k is predicted based on the optimal estimate at time k-1; P k-1 and P k They are the posterior estimated covariances at time k-1 and time k (i.e. x k-1 and x k covariance of P k is the prior estimated covariance (x k covariance of K k is the Kalman gain.

[0076] Further, refer to Figure 2 , the step S220 includes:

[0077] S221, first use the Kalman filter to recursively estimate the state from the initial moment to the final moment;

[0078] S222: After the forward filtering is completed, the obtained Kalman filtering result is used to perform backward smoothing;

[0079] S223. Recursively calculate from the last moment forward, adjust the state estimate at each moment, and update the state at the previous moment in combination with the state information at the subsequent moment to further reduce the error.

[0080] Further, refer to Figure 2 In step S230,

[0081] Using the location information provided by the GPS device, the driving speed is calculated in real time. Based on the Haversine formula, the surface distance d between two points is calculated. Assume that two consecutive location points P1 and P2 are obtained from the GPS device. The two location points include longitude and latitude coordinates and timestamps P1 (longitude λ1, latitude φ1, time t1) and P2 (longitude λ2, latitude φ2, time t2). Substitute into:

[0082]

[0083] Where φ and λ are in radians, R is the radius of the Earth, d is the distance between two points, and velocity v is the rate of change of distance over time. The formula for calculating velocity v is:

[0084]

[0085] Based on the data provided by the GPS device, the vehicle's acceleration value per second is calculated in real time. Severe negative acceleration indicates that the vehicle has braked suddenly or collided. Three consecutive positions P1, P2, and P3 are obtained from the GPS device, corresponding to time points t1, t2, and t3, respectively. First, the velocity v formula is applied to calculate the velocity v1 from P1 to P2 and the velocity v2 from P2 to P3. Then, the acceleration a is calculated:

[0086]

[0087] The angle α between the initial acceleration and the acceleration during driving is:

[0088]

[0089] in, is the initial acceleration value when the vehicle starts, is the acceleration value of the vehicle during driving;

[0090] In step S230, the position changes of the vehicle are monitored in real time through the GPS device data. If the vehicle suddenly stops during driving and the position hardly changes, the speed and acceleration of the vehicle are combined to assist in determining whether a collision has occurred. Collisions usually occur in a very short period of time. By monitoring the time interval of speed changes, it is assisted in determining whether a collision has occurred.

[0091] Further, refer to Figure 2 The angle α between the initial acceleration and the acceleration during driving is in the range of 40-55 degrees.

[0092] Further, refer to Figure 3 , the step S300 includes:

[0093] S310, using an on-board camera image to identify the driving environment and weather conditions, and identifying the vehicle's driving state based on the vehicle's speed, acceleration, GPS coordinates, and steering angle, where the driving environment includes at least the GPS location, and the vehicle's driving state includes at least normal driving, sudden acceleration, sudden deceleration, and sharp turn;

[0094] S320, analyzing the current driving environment, weather conditions, and vehicle driving status through a deep learning model, and dynamically adjusting collision detection thresholds and parameters;

[0095] S330. Update parameter settings based on real-time data to optimize collision detection performance, and adjust thresholds based on detection results to reduce false positives and false negatives.

[0096] Further, refer to Figure 3 In step S320,

[0097] The deep learning model includes a lightweight neural network model MobileNet and a long short-term memory network LSTM;

[0098] The neural network model MobileNet is used to identify driving environment and weather conditions from images from the vehicle camera, and the long short-term memory network LSTM is used to analyze the vehicle driving status.

[0099] Further, refer to Figure 4 The present invention further proposes a collision detection device based on multi-sensor data fusion, which is used to execute the collision detection method based on multi-sensor data fusion. The collision detection device based on multi-sensor data fusion includes:

[0100] a data acquisition unit, the data acquisition unit comprising at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting a steering angle of a steering wheel;

[0101] A data processing unit, which is based on a SOC (System on Chip) platform and includes at least a central processing unit (CPU) and a neural network processor (NPU);

[0102] The alarm and recording unit and the data processing unit perform collision detection, and then generate alarm information to the board and mobile APP at the same time, and save the video data to the local storage card or cloud space.

[0103] Furthermore, the present invention also proposes a computer-readable storage medium having program instructions stored thereon, and when the program instructions are executed by a processor, the collision detection method based on multi-sensor data fusion is implemented.

[0104] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the aforementioned embodiments. As long as the technical effects of the present invention are achieved by the same means, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure shall be included within the scope of protection of the present disclosure. Within the scope of protection of the present invention, its technical solutions and / or implementation methods may be modified and varied in various ways.

Claims

1. A collision detection method based on multi-sensor data fusion, wherein the collision detection method based on multi-sensor data fusion is applied to a collision detection device based on multi-sensor data fusion, wherein the collision detection device based on multi-sensor data fusion comprises a data acquisition unit, wherein the data acquisition unit comprises at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting the steering angle of the steering wheel; a data processing unit, wherein the data processing unit is based on a system-on-chip SOC platform and comprises at least a central processing unit CPU and a neural network processor NPU; an alarm and recording unit, wherein the data processing unit performs collision detection, and then generates alarm information to the board end and the mobile end APP at the same time, and saves the video data to a local storage card or a cloud space at the same time, wherein the data processing unit performs collision detection, and then generates alarm information to the board end and the mobile end APP at the same time, and saves the video data to a local storage card or a cloud space, wherein the data processing unit performs collision detection, and then generates alarm information to the board end and the mobile end APP at the same time, and saves the video data to a local storage card or a cloud space, wherein the data processing unit performs collision detection, and then generates alarm information to the board end and the mobile end APP at the same time, and saves the video data to a local storage card or a cloud space, and ... The collision detection method based on multi-sensor data fusion comprises the following steps: S100, collecting dynamic information of the vehicle in real time through multiple sensors installed on the vehicle; S200, preprocessing the dynamic information collected by multiple sensors and saving the preprocessed data; S300, automatically adjusting the collision detection strategy and threshold of the deep learning model according to the real-time driving environment and vehicle status of the vehicle; S400, comprehensive analysis and collision detection, based on the parameters and threshold results adjusted in real time, the pre-processed sensor data is comprehensively analyzed to determine the collision detection conditions. When it is detected that the acceleration, angular velocity and steering angle meet the collision threshold, the system determines that a collision has occurred; S500, once a collision is detected, automatically save video records, sensor data and related parameters before and after the collision.

2. The collision detection method based on multi-sensor data fusion according to claim 1, characterized in that: In the step S100, The multiple sensors include at least a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting the steering angle of the steering wheel.

3. The collision detection method based on multi-sensor data fusion according to claim 1, characterized in that: The step S200 includes: S210, synchronously aligning the timestamps of the data of each sensor to ensure that the data are consistent in time; S220, using Kalman filter and Gaussian smoothing to denoise multiple sets of sensor data and filter out data abnormalities; S230, performing calculations based on the filtered data, including vehicle speed, acceleration, angle between initial acceleration and acceleration during driving, position change and time interval, and saving the calculation results in chronological order.

4. The collision detection method based on multi-sensor data fusion according to claim 3 is characterized in that: The step S220 includes: S221, first use the Kalman filter to recursively estimate the state from the initial moment to the final moment; S222, after the forward filtering is completed, using the obtained Kalman filtering result to perform backward smoothing; S223. Recursively calculate from the last moment forward, adjust the state estimate at each moment, and update the state at the previous moment in combination with the state information at the subsequent moment to further reduce the error.

5. The collision detection method based on multi-sensor data fusion according to claim 3, characterized in that: In the step S230, Using the location information provided by the GPS device, the driving speed is calculated in real time. Based on the Haversine formula, the surface distance d between two points is calculated. Assume that two consecutive location points P1 and P2 are obtained from the GPS device. The two location points include longitude and latitude coordinates and timestamps P1 (longitude λ1, latitude φ1, time t1) and P2 (longitude λ2, latitude φ2, time t2). Substitute into: d=R·c Among them, φ and λ are in radians, R is the radius of the earth, d is the distance between two points, and velocity v is the rate of change of distance over time. The calculation formula for velocity v is: According to the data provided by the GPS device, the acceleration value of the vehicle per second is calculated in real time. The severe negative acceleration indicates that the vehicle has braked suddenly or collided. Three consecutive position points P1, P2 and P3 are obtained from the GPS device, corresponding to the time points t1, t2 and t3 respectively. First, the speed v formula is applied to calculate the speed v1 from P1 to P2 and the speed v2 from P2 to P3, and then the acceleration a is calculated: The angle α between the initial acceleration and the acceleration during driving is: in, is the initial acceleration value when the vehicle starts. is the acceleration value of the vehicle during driving; In step S230, the position change of the vehicle is monitored in real time through the GPS device data. If the vehicle suddenly stops during driving and the position hardly changes, the speed and acceleration of the vehicle are combined to assist in determining whether a collision has occurred. Collisions usually occur in a very short period of time. By monitoring the time interval of speed changes, it is assisted in determining whether a collision has occurred.

6. The collision detection method based on multi-sensor data fusion according to claim 5, characterized in that: The angle α between the initial acceleration and the acceleration during driving is in the range of 40-55 degrees.

7. The collision detection method based on multi-sensor data fusion according to claim 1, characterized in that: The step S300 includes: S310, using the vehicle camera image to identify the driving environment and weather conditions, and identifying the vehicle driving state through the vehicle speed, acceleration, GPS coordinates and steering angle, wherein the driving environment at least includes the GPS position and the vehicle driving state at least includes normal driving, rapid acceleration, rapid deceleration and sharp turn; S320, analyzing the current driving environment, weather conditions, and vehicle driving status through a deep learning model, and dynamically adjusting the thresholds and parameters of collision detection; S330, update parameter settings according to real-time data, optimize collision detection performance, and adjust thresholds according to detection results to reduce false positives and false negatives.

8. The collision detection method based on multi-sensor data fusion according to claim 7, characterized in that: In the step S320, The deep learning model includes a lightweight neural network model MobileNet and a long short-term memory network LSTM; The neural network model MobileNet is used to identify driving environment and weather conditions from images from the vehicle-mounted camera, and the long short-term memory network LSTM is used to analyze the vehicle driving status.

9. A collision detection device based on multi-sensor data fusion, used to execute the collision detection method based on multi-sensor data fusion as claimed in any one of claims 1 to 8, characterized in that: The collision detection device based on multi-sensor data fusion includes: A data acquisition unit, the data acquisition unit at least comprising a three-axis accelerometer for detecting vehicle acceleration, a three-axis gyroscope for detecting vehicle angular velocity, a GPS device for providing geographic location information and speed data of the vehicle, and a steering angle sensor for detecting a steering angle of a steering wheel; A data processing unit, which is based on a system-on-chip (SOC) platform and includes at least a central processing unit (CPU) and a neural network processor (NPU); The alarm and recording unit and the data processing unit perform collision detection, and then generate alarm information to the board and mobile APP at the same time, and save the video data to the local storage card or cloud space. 10 . A computer-readable storage medium having program instructions stored thereon, wherein the program instructions are executed by a processor to implement the method according to claim 1 .

Images