A method for determining traffic accidents

Abstract

This invention belongs to the field of traffic accident determination, specifically relating to a traffic accident determination method, including the following steps: S1, acquiring vehicle parameters; S2, calculating the feature values ​​of the vehicle parameters; S3, forming several vehicle driving models based on the feature values ​​and a preset algorithm; S4, determining whether the current state of the vehicle satisfies any of the vehicle driving models, and if not, determining that the vehicle has been involved in a traffic accident. This invention solves the problems of existing traffic accident determination technologies, which partially rely solely on single IMU data, making them susceptible to sensor drift and road surface disturbances, resulting in a high false positive rate; furthermore, the determination method based on a combination of sensor data and data thresholds uses a unified determination model, which cannot accommodate differences in sensor data under different vehicle conditions, leading to poor versatility and reliability, and failing to meet the needs of accident determination for multiple vehicle models and conditions.

Description

Technical Field

[0001] This invention belongs to the field of traffic accident determination, and specifically relates to a method for determining traffic accidents. Background Technology

[0002] Currently, in-vehicle terminals are increasingly integrating sensors such as inertial measurement units (IMUs) and global positioning systems (GPS), enabling real-time collection of multi-dimensional driving data, including vehicle acceleration, angular velocity, position, and speed. This provides a data foundation for intelligent post-accident accident assessment. Currently, accident assessment often uses pre-set models. For example, a vibration value model is established based on IMU data; when the vehicle's vibration value exceeds a preset threshold, it is considered an accident. Alternatively, data from various scenarios, including normal driving and collisions, is collected, labeled, and used to establish a classification model.

[0003] Existing traffic accident determination technologies, some of which rely solely on single IMU data, are susceptible to interference from sensor drift and road surface bumps, leading to a high rate of false judgments. In addition, determination methods based on a combination of sensor data and data thresholds use a unified determination model, which cannot be compatible with the differences in sensor data under different vehicle conditions and vehicle models, resulting in poor versatility and reliability, and failing to meet the needs of accident determination under multiple vehicle models and conditions.

[0004] Therefore, this invention proposes a method for determining traffic accidents. Summary of the Invention

[0005] The purpose of this invention is to disclose a traffic accident determination method that solves the problems of existing traffic accident determination technologies, which rely on a single IMU data and are easily affected by sensor drift and road surface bumps, resulting in a high misjudgment rate. In addition, the determination method based on the combination of sensor data and data thresholds adopts a unified determination model, which cannot be compatible with the differences in sensor data under different vehicle conditions, resulting in poor versatility and reliability, and failing to meet the needs of accident determination under multiple vehicle models and conditions.

[0006] To achieve the above objectives, this invention discloses a method for determining traffic accidents, comprising the following steps: S1, Obtain vehicle parameters; S2, calculate the characteristic values ​​of the vehicle parameters; S3, Several vehicle driving models are formed based on the feature values ​​and preset algorithms; S4. Determine whether the current state of the vehicle satisfies any of the vehicle driving models described above; otherwise, determine that the vehicle has been involved in a traffic accident.

[0007] As an optional implementation, the vehicle parameters in step S1 include speed, acceleration, and angular velocity; The acceleration includes the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration in a spatial coordinate system with the vehicle center as the origin, the travel direction as the X-axis, and the vehicle height direction as the Y-axis. The angular velocity includes the X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity in the spatial coordinate system.

[0008] As an optional implementation, the vehicle parameters in step S1 are stored to form dataset A. ; ; Among them, A t The data generated in second t represents the total number of seconds the data is stored; Time represents the number of data samples per second; fps represents the vehicle speed; xAcc represents the X-axis acceleration; yAcc represents the Y-axis acceleration; zAcc represents the Z-axis acceleration; xGyr represents the X-axis angular velocity; yGyr represents the Y-axis angular velocity; and zGyr represents the Z-axis angular velocity.

[0009] As an optional implementation, step S2 includes: Multidimensional feature value calculation is performed on the dataset A to obtain feature set B. ; in, Represents the dynamic characteristic values ​​of vehicle parameters.

[0010] As an optional implementation, the dynamic feature value This includes vehicle parameters and processed data after processing the vehicle parameters; the processed data includes at least one of the following: mean, variance, Manhattan norm, and maximum neighbor difference. ; ; ; ; Wherein, data represents any data in the vehicle parameters, including speed, acceleration, and angular velocity; μ represents the mean of any data in the vehicle parameters; var represents the variance of any data in the vehicle parameters; S represents the Manhattan norm of any data in the vehicle parameters; and δ represents the maximum adjacent difference of any data in the vehicle parameters.

[0011] As an optional implementation, step S3 includes: S31, use a clustering algorithm to divide the feature set B into k categories and obtain the cluster center set C; , ; Among them, C i (i=1,2,...,k) represents the i-th cluster center of feature set B; S32, Calculate the covariance matrix of the k classes. , , ; in, This represents the sample data of the i-th category. This represents the number of samples in the i-th category; S33, Calculate the equation of the multidimensional hyperellipsoid. ; in, This represents the scaling factor of the hyperellipsoid.

[0012] As an optional implementation, step S4 includes: S41, Calculate the multidimensional feature values ​​of the vehicle parameters acquired in real time; S42. If the calculated multidimensional eigenvalues ​​satisfy any ellipsoid equation, the vehicle is determined to be driving normally; otherwise, a traffic accident is determined to have occurred.

[0013] As an optional implementation, the vehicle parameters in step S1 are obtained by accessing the vehicle body sensors.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: First, the traffic accident determination method proposed in this invention constructs a vehicle driving model of the vehicle based on the vehicle parameters of the actual vehicle, thereby achieving adaptive construction and matching, improving the adaptability of the traffic accident determination method to the vehicle, and thus improving the accuracy of the judgment results, which is helpful for subsequent accident determination, insurance compensation and other matters. Second, the traffic accident determination method proposed in this invention obtains vehicle parameters by observing the actual driving state of the vehicle. Based on the actual road conditions and driving environment, a driving model that can fully cover all vehicles under normal driving conditions can be constructed to ensure the accuracy of subsequent state determination. Third, the traffic accident determination method proposed in this invention removes edge parameters and abrupt parameters at the time of the accident from all the acquired vehicle parameters through a clustering algorithm, so that the clustered vehicle driving models are all vehicle driving models under normal driving conditions. Fourth, the traffic accident determination method proposed in this invention obtains the vehicle driving model through a clustering algorithm, which does not require training and has low computational complexity. It can be generated when the vehicle driving time reaches the minimum threshold, thereby improving the efficiency of accident determination. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] The technical solution of the present invention will be further described below with reference to the embodiments.

[0017] This embodiment proposes a method for determining traffic accidents, including the following steps: S1, Obtain vehicle parameters; S2, calculates the characteristic values ​​of the vehicle parameters; S3, Several vehicle driving models are formed based on feature values ​​and preset algorithms; S4 determines whether the current state of the vehicle satisfies any vehicle driving model; otherwise, it determines that the vehicle has been involved in a traffic accident.

[0018] In a preferred embodiment, the vehicle parameters in step S1 are obtained by calling on-board sensors, typically by calling the inertial measurement unit of the on-board terminal.

[0019] In a preferred embodiment, steps S1 to S3 are executed during the vehicle testing phase or the put-in-use phase, or they can be executed after each power-on use of the vehicle; step S4 can be executed synchronously with steps S1 to S3. During the continuous acquisition process of steps S1 to S3, step S4 compares the standard with the pre-formed vehicle driving model periodically.

[0020] In this embodiment, since the vehicle is mostly in normal driving mode and traffic accidents account for a small proportion, the vehicle driving model obtained in steps S1 to S3 is the normal driving model under different vehicle states. When a traffic accident occurs, the vehicle parameters generated by the collision will change abruptly, which is much greater than that of normal driving. Therefore, it can be distinguished from the normal driving model to ensure the accurate determination of traffic accidents. At the same time, real vehicle testing is carried out for different vehicle models.

[0021] In step S1, the vehicle parameters include speed, acceleration, and angular velocity. Specifically, spatial coordinates are established with the vehicle center as the origin, the direction of travel as the X-axis, and the vehicle height direction as the Y-axis. The acceleration in the vehicle parameters includes X-axis acceleration, Y-axis acceleration, and Z-axis acceleration; the angular velocity in the vehicle parameters includes X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity.

[0022] For example, vehicle parameters are stored to form dataset A. ; ; Among them, A t The data generated in second t represents the total number of seconds the data is stored; Time represents the number of data samples per second; fps represents the vehicle speed; xAcc represents the X-axis acceleration; yAcc represents the Y-axis acceleration; zAcc represents the Z-axis acceleration; xGyr represents the X-axis angular velocity; yGyr represents the Y-axis angular velocity; and zGyr represents the Z-axis angular velocity.

[0023] In a preferred embodiment, in step S1, if vehicle positioning data cannot be obtained, it may be due to a vehicle sensor malfunction, in which case the obtained vehicle parameters are inaccurate and therefore not obtained; it may also be due to the vehicle being in a tunnel, basement, or other road section, where the vehicle's driving state is not a normal driving state and therefore not obtained.

[0024] In this embodiment, obtaining the vehicle's speed, acceleration, and angular velocity helps to compare the vehicle's state with the actual state when a traffic accident occurs. For example, when an accident occurs, the acceleration can reach several g or even tens of g, which is far from the ±0.5g of normal driving. The speed and angular velocity are also similar. Therefore, by obtaining the normal values ​​of speed, acceleration, and angular velocity, it is possible to determine whether a traffic accident has occurred.

[0025] Step S2 includes: Multidimensional feature value calculation is performed on the dataset A to obtain feature set B. ; in, Represents the dynamic characteristic values ​​of vehicle parameters.

[0026] Furthermore, dynamic eigenvalues This includes, but is not limited to, any data such as speed, acceleration, and angular velocity, nor is it limited to processed data such as mean, variance, Manhattan norm, and maximum adjacent difference. For example, let 'data' represent any data such as speed, acceleration, and angular velocity in the vehicle parameters; the mean of any data is calculated as follows: ; The variance of any data point is calculated as follows: ; The Manhattan norm is calculated for any data as follows: ; The maximum adjacent difference is calculated for any data as follows: .

[0027] Specifically, step S3 includes: S31, use a clustering algorithm to divide the feature set B into k categories and obtain the cluster center set C; , ; Among them, C i (i=1,2,...,k) represents the i-th cluster center of feature set B; S32, Calculate the covariance matrix of the k classes. , , ; in, This represents the sample data of the i-th category. This represents the number of samples in the i-th category; S33, Calculate the equation of the multidimensional hyperellipsoid. ; in, This represents the scaling factor of the hyperellipsoid.

[0028] In a preferred embodiment, step S31 further employs the elbow rule, calling the K-Means++ clustering algorithm to divide the feature set B into k categories.

[0029] In a preferred embodiment, step S33 takes... =3, which can include 99.7% of the samples, while When =1, it contains 68% of the sample size, while When the value is 2, it contains 95% of the sample size, that is... The larger the value, the lower the false alarm rate. The smaller the value, the higher the recall rate.

[0030] Additionally, step S4 includes: S41, calculate the multidimensional feature values ​​of the vehicle parameters acquired in real time; S42. If the calculated multidimensional eigenvalues ​​satisfy any ellipsoid equation, the vehicle is determined to be driving normally; otherwise, a traffic accident is determined to have occurred.

[0031] In a preferred embodiment, the vehicle parameters in step S41 include speed, triaxial acceleration, and triaxial angular velocity.

[0032] In a preferred embodiment, step S42 calculates whether the multidimensional eigenvalues ​​satisfy any ellipsoid equation, and verifies them sequentially.

[0033] In this embodiment, modeling is based on the data of each vehicle, which is simple, easy to implement, and more accurate. At the same time, it does not require prior data collection, saving a lot of time and costs. Furthermore, the vehicle driving model is generated through clustering algorithms, which reduces computational complexity. Also, the accuracy of traffic accident judgment can be guaranteed without prior model training.

[0034] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for determining traffic accidents, characterized in that, Includes the following steps: S1, Obtain vehicle parameters; S2, calculate the characteristic values ​​of the vehicle parameters; S3, Several vehicle driving models are formed based on the feature values ​​and preset algorithms; S4. Determine whether the current state of the vehicle satisfies any of the vehicle driving models described above; otherwise, determine that the vehicle has been involved in a traffic accident.

2. The method for determining a traffic accident according to claim 1, characterized in that: The vehicle parameters in step S1 include speed, acceleration, and angular velocity; The acceleration includes the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration in a spatial coordinate system with the vehicle center as the origin, the travel direction as the X-axis, and the vehicle height direction as the Y-axis. The angular velocity includes the X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity in the spatial coordinate system.

3. The method for determining a traffic accident according to claim 2, characterized in that: The vehicle parameters in step S1 are stored to form dataset A. ； ； Among them, A t The data generated in second t represents the total number of seconds the data is stored; Time represents the number of data samples per second; fps represents the vehicle speed; xAcc represents the X-axis acceleration; yAcc represents the Y-axis acceleration; zAcc represents the Z-axis acceleration; xGyr represents the X-axis angular velocity; yGyr represents the Y-axis angular velocity; and zGyr represents the Z-axis angular velocity.

4. The method for determining a traffic accident according to claim 3, characterized in that, Step S2 includes: Multidimensional feature value calculation is performed on the dataset A to obtain feature set B. ； in, Represents the dynamic characteristic values ​​of vehicle parameters.

5. The method for determining a traffic accident according to claim 4, characterized in that: The dynamic eigenvalues This includes vehicle parameters and processed data after processing the vehicle parameters; the processed data includes at least one of the following: mean, variance, Manhattan norm, and maximum neighbor difference. ； ； ； ； Wherein, data represents any data in the vehicle parameters, including speed, acceleration, and angular velocity; μ represents the mean of any data in the vehicle parameters; var represents the variance of any data in the vehicle parameters; S represents the Manhattan norm of any data in the vehicle parameters; and δ represents the maximum adjacent difference of any data in the vehicle parameters.

6. The method for determining a traffic accident according to claim 4, characterized in that, Step S3 includes: S31, use a clustering algorithm to divide the feature set B into k categories and obtain the cluster center set C; ， ； Among them, C i (i=1,2,...,k) represents the i-th cluster center of feature set B; S32, Calculate the covariance matrix of the k classes. , ， ； in, This represents the sample data of the i-th category. This represents the number of samples in the i-th category; S33, Calculate the equation of the multidimensional hyperellipsoid. ； in, This represents the scaling factor of the hyperellipsoid.

7. The method for determining a traffic accident according to claim 6, characterized in that, Step S4 includes: S41, Calculate the multidimensional feature values ​​of the vehicle parameters acquired in real time; S42. If the calculated multidimensional eigenvalues ​​satisfy any ellipsoid equation, the vehicle is determined to be driving normally; otherwise, a traffic accident is determined to have occurred.

8. A method for determining traffic accidents according to any one of claims 1 to 7, characterized in that: The vehicle parameters in step S1 are obtained by accessing the vehicle body sensors.

Images