Vehicle accident detection method and device, storage medium, vehicle and monitoring end

Abstract

The application provides a vehicle accident detection method and device, a storage medium, a vehicle and a monitoring end, and belongs to the technical field of vehicles. According to the application, the Internet of Vehicles sequence data of a target vehicle is acquired, and the Internet of Vehicles sequence data is input into at least one target statistical model, so that at least one abnormal data detection result can be obtained. Meanwhile, the Internet of Vehicles sequence data is input into a pre-trained deep learning model, so that an accident detection result can be obtained. Then, based on the at least one abnormal data detection result and the accident detection result, it can be determined whether the target vehicle has an accident. According to the application, the target statistical model and the deep learning model are jointly used, the Internet of Vehicles sequence data can be analyzed from multiple angles, a large amount of Internet of Vehicles data with various data types can be fully utilized, the robustness of the algorithm is improved, and the accuracy of vehicle accident detection is effectively improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a vehicle accident detection method, device, storage medium, vehicle, and monitoring terminal. Background Technology

[0002] With the development of the Internet of Vehicles, it has become increasingly easy to acquire various types of vehicle data. How to use Internet of Vehicles data for vehicle accident detection has become an important research direction.

[0003] Currently, vehicle accident detection is typically based on a single data source or traditional machine learning algorithms, such as decision trees and support vector machines. However, due to the complexity of vehicle driving states and environments, single detection methods struggle to fully utilize the large volume and diverse types of vehicle network data, resulting in low accuracy in identifying vehicle accidents. Summary of the Invention

[0004] This application provides a vehicle accident detection method, device, storage medium, vehicle, and monitoring terminal to solve the problem that current single detection methods cannot fully utilize the large amount and diverse types of vehicle network data, resulting in a low accuracy rate in identifying vehicle driving accidents.

[0005] To solve the above problems, this application adopts the following technical solution:

[0006] In a first aspect, embodiments of this application provide a vehicle accident detection method, the method comprising:

[0007] Obtain the vehicle-to-everything (V2X) sequence data of the target vehicle;

[0008] The vehicle network sequence data is input into at least one pre-trained target statistical model to obtain at least one abnormal data detection result; different target statistical models are used to detect abnormal driving data of different types.

[0009] The vehicle network sequence data is input into a pre-trained deep learning model to obtain accident detection results;

[0010] Based on the at least one abnormal data detection result and the accident detection result, it is determined whether the target vehicle has been involved in a driving accident.

[0011] In one embodiment of this application, the at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; the at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result;

[0012] The step of inputting the vehicle network sequence data into at least one target statistical model to obtain at least one abnormal data detection result includes:

[0013] Based on the vehicle network sequence data, acceleration sequence data, steering wheel speed sequence data, and / or parking behavior sequence data are obtained;

[0014] The acceleration sequence data is input into the acceleration anomaly detection model, and the acceleration anomaly detection result is output; and / or,

[0015] The steering wheel speed sequence data is input into the steering wheel speed anomaly detection model, and the steering wheel speed anomaly detection result is output; and / or,

[0016] The parking behavior sequence data is input into the parking behavior anomaly detection model, and the parking behavior anomaly detection result is output.

[0017] In one embodiment of this application, the step of inputting the vehicle network sequence data into a deep learning model to obtain accident detection results includes:

[0018] Based on the vehicle network sequence data, vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data are obtained;

[0019] The vehicle speed sequence data, the acceleration sequence data, and the steering wheel rotation speed sequence data are input into the deep learning model to obtain the accident detection results.

[0020] In one embodiment of this application, the step of determining whether the target vehicle has been involved in a driving accident based on the at least one abnormal data detection result and the accident detection result includes:

[0021] The first detection time for the acceleration anomaly detection result, the second detection time for the steering wheel speed anomaly detection result, the third detection time for the parking behavior anomaly detection result, and the fourth detection time for the accident detection result are determined.

[0022] If the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions, based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, it is determined whether the target vehicle has been involved in a driving accident.

[0023] In one embodiment of this application, before the step of determining whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, when the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions, the method further includes:

[0024] If the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time, then the first detection time, the second detection time, the third detection time and the fourth detection time are determined to satisfy the preset condition.

[0025] In one embodiment of this application, the step of determining whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result includes:

[0026] The accident score is obtained by weighted summation of the acceleration anomaly detection results, the steering wheel speed anomaly detection results, the parking behavior anomaly detection results, and the accident detection results.

[0027] If the accident score is greater than the score threshold, it is determined that the target vehicle has been involved in a driving accident.

[0028] In one embodiment of this application, the at least one target statistical model and the deep learning model are trained in the following manner:

[0029] For any of the target statistical models, based on the target data type of the abnormal driving data detected by the target statistical model, the target data sequence of the target data type is extracted from the historical vehicle network sequence data;

[0030] Accident tags are labeled on the target data sequence and the historical vehicle network sequence data to obtain the first model training samples and the second model training samples, respectively; the accident tags include whether an accident has occurred and the time of the accident.

[0031] The initial statistical model is trained using the training samples of the first model to obtain the target statistical model;

[0032] The initial deep learning model is trained using the training samples of the second model to obtain the deep learning model.

[0033] Secondly, based on the same inventive concept, embodiments of this application provide a vehicle accident detection device, the device comprising:

[0034] The data acquisition module is used to acquire the vehicle-to-everything (V2X) sequence data of the target vehicle.

[0035] The first detection module is used to input the vehicle network sequence data into at least one target statistical model to obtain at least one abnormal data detection result; different target statistical models are used to detect abnormal driving data of different types.

[0036] The second detection module is used to input the vehicle network sequence data into a deep learning model to obtain accident detection results;

[0037] An accident determination module is used to determine whether the target vehicle has been involved in a driving accident based on the at least one abnormal data detection result and the accident detection result.

[0038] In one embodiment of this application, the at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; the at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result; the first detection module includes:

[0039] The first data extraction submodule is used to obtain acceleration sequence data, steering wheel speed sequence data and / or parking behavior sequence data based on the vehicle network sequence data;

[0040] An acceleration detection submodule is used to input the acceleration sequence data into the acceleration anomaly detection model and output the acceleration anomaly detection result.

[0041] The steering wheel speed detection submodule is used to input the steering wheel speed sequence data into the steering wheel speed anomaly detection model and output the steering wheel speed anomaly detection result.

[0042] The parking behavior detection submodule is used to input the parking behavior sequence data into the parking behavior anomaly detection model and output the parking behavior anomaly detection result.

[0043] In one embodiment of this application, the second detection module includes:

[0044] The second data extraction submodule is used to obtain vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data based on the vehicle network sequence data;

[0045] The accident detection submodule is used to input the vehicle speed sequence data, the acceleration sequence data, and the steering wheel rotation speed sequence data into the deep learning model to obtain the accident detection result.

[0046] In one embodiment of this application, the accident determination module includes:

[0047] The detection time determination submodule is used to determine the first detection time of the acceleration anomaly detection result, the second detection time of the steering wheel speed anomaly detection result, the third detection time of the parking behavior anomaly detection result, and the fourth detection time of the accident detection result.

[0048] The accident determination submodule is used to determine whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, provided that the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions.

[0049] In one embodiment of this application, the vehicle accident detection device further includes:

[0050] The condition determination module is used to determine that the first detection time, the second detection time, the third detection time, and the fourth detection time satisfy the preset condition when the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time.

[0051] In one embodiment of this application, the accident determination submodule includes:

[0052] The weighted summation unit is used to perform a weighted summation of the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result to obtain an accident score value.

[0053] An accident determination unit is used to determine that the target vehicle has been involved in a driving accident when the accident score value is greater than a score threshold.

[0054] In one embodiment of this application, the vehicle accident detection device further includes a model training module, the model training module comprising:

[0055] The third data extraction submodule is used to extract the target data sequence of the target data type from historical vehicle network sequence data based on the target data type of the abnormal driving data detected by the target statistical model for any of the target statistical models.

[0056] The data labeling submodule is used to label the target data sequence and the historical vehicle network sequence data with accident tags, thereby obtaining the first model training samples and the second model training samples, respectively; the accident tags include whether an accident has occurred and the time of the accident.

[0057] The first model training submodule is used to train the initial statistical model using the first model training samples to obtain the target statistical model;

[0058] The second model training submodule is used to train the initial deep learning model using the training samples of the second model to obtain the deep learning model.

[0059] Thirdly, based on the same inventive concept, embodiments of this application provide a storage medium storing machine-executable instructions, which, when executed by a processor, implement the vehicle accident detection method proposed in the first aspect of this application.

[0060] Fourthly, based on the same inventive concept, embodiments of this application provide a vehicle including a vehicle accident detection system, the vehicle accident detection system being used to execute the vehicle accident detection method proposed in the first aspect of this application.

[0061] Fifthly, based on the same inventive concept, embodiments of this application provide a monitoring terminal, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor is used to execute the machine-executable instructions to implement the vehicle accident detection method proposed in the first aspect of this application.

[0062] Compared with the prior art, this application has the following advantages:

[0063] This application provides a vehicle accident detection method. By acquiring vehicle-to-everything (V2X) sequence data of a target vehicle and inputting this data into at least one target statistical model, at least one abnormal data detection result can be obtained. Simultaneously, by inputting the V2X sequence data into a pre-trained deep learning model, an accident detection result can be obtained. Based on both the abnormal data detection result and the accident detection result, it is possible to determine whether the target vehicle has been involved in a driving accident. This application, through the target statistical model, can effectively identify abnormal driving data of various data types within the V2X sequence data. Furthermore, by combining the deep learning model, accident analysis is performed on the V2X sequence data from multiple perspectives, achieving full utilization of the large amount of V2X data with diverse data types. This improves the robustness of the algorithm and effectively enhances the accuracy of vehicle accident detection. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a schematic diagram of the steps of a vehicle accident detection method in one embodiment of this application.

[0066] Figure 2 This is a schematic diagram of the functional modules of a vehicle accident detection device according to one embodiment of this application.

[0067] Figure 3 This is a schematic diagram of the structure of a monitoring terminal in one embodiment of this application. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] It's important to note that vehicle-to-everything (V2X) data is typically composed of time-series data, encompassing massive amounts of diverse vehicle driving data. Current vehicle accident detection methods are usually based on a single data source or traditional machine learning algorithms. These algorithms struggle to fully utilize the time-series data from V2X or cannot handle large-scale datasets. Furthermore, due to the complexity of vehicle driving states and environments, a single detection method may not accurately detect accidents. For example, relying solely on anomaly detection based on acceleration or steering wheel speed might overlook other crucial information, such as driver actions.

[0070] Therefore, when faced with large-scale vehicle network data, a real-time and efficient vehicle accident detection method is needed, which can comprehensively analyze various vehicle network data in a timely and accurate manner to improve the accuracy of vehicle driving accident detection.

[0071] To address the problem that current single detection methods struggle to fully utilize the large volume and diverse types of vehicle network data, resulting in low accuracy in identifying vehicle accidents, this application aims to provide a vehicle accident detection method. By combining at least one target statistical model and a deep learning model, accident analysis is performed on vehicle network sequence data from multiple perspectives, enabling full utilization of the vast amount of diverse vehicle network data. This improves both the robustness of the algorithm and the accuracy of vehicle accident detection.

[0072] Reference Figure 1 This application illustrates a vehicle accident detection method, which may include the following steps:

[0073] S101: Obtain the vehicle network sequence data of the target vehicle.

[0074] It should be noted that all or part of the steps in this embodiment can be executed either on the target vehicle's local electronic equipment or on a server or other remote device with communication and / or data processing capabilities with the vehicle, depending on actual needs. The server can be a central server, a cluster server, or a distributed server, or a cloud server implementing cloud computing and / or cloud storage, etc. It should be noted that this embodiment does not impose specific restrictions on the executing entity.

[0075] Specifically, when the executing entity is a server, the server communicates with at least one target vehicle and acquires the vehicle network sequence data of each target vehicle in real time to realize vehicle accident detection for any target vehicle; when the executing entity is a local electronic device, the local electronic device will collect the vehicle network sequence data of the target vehicle in real time, perform real-time detection of vehicle accidents locally, and upload the detection results to the server when a driving accident is detected.

[0076] In this embodiment, the vehicle-to-everything (V2X) sequence data refers to vehicle driving data generated by the target vehicle according to a time sequence. Specifically, it can include vehicle driving data of various data types such as vehicle speed, acceleration, steering wheel speed, and parking behavior. For example, for a specific moment in the vehicle's driving process, the corresponding vehicle driving data of various data types such as vehicle speed, acceleration, steering wheel speed, and parking behavior can be retrieved based on the V2X sequence data. For a specific time period in the vehicle's driving process, the corresponding vehicle speed sequence data, acceleration sequence data, steering wheel speed sequence data, and parking behavior sequence data can be obtained. The data interval of the V2X sequence data can be on the order of seconds.

[0077] S102: Input the vehicle network sequence data into at least one pre-trained target statistical model to obtain at least one abnormal data detection result.

[0078] In practical implementation, since the vehicle network sequence data is time series data, the target statistical model can adopt a statistical model used for time series analysis, such as the autoregressive moving average model, also known as the ARMA model.

[0079] In this embodiment, abnormal driving data in vehicle network sequence data can be effectively detected by using a pre-trained target statistical model to identify abnormal data in the vehicle network sequence data.

[0080] It should be noted that different target statistical models are used to detect abnormal driving data of different types. For example, the target statistical model may include an acceleration anomaly detection model and / or a steering wheel speed anomaly detection model, wherein the acceleration anomaly detection model is used to detect abnormal acceleration data in vehicle network sequence data, while the steering wheel speed anomaly detection model is used to detect abnormal steering wheel speed data in vehicle network sequence data.

[0081] In this embodiment, by setting corresponding target statistical models for different types of vehicle network sequence data, it is possible to effectively detect various abnormal driving data from different dimensions.

[0082] S103: Input the vehicle network sequence data into the pre-trained deep learning model to obtain the accident detection results.

[0083] In this embodiment, although the target statistical model can effectively detect abnormal driving data, the occurrence of abnormal driving data does not necessarily indicate the occurrence of an accident. For example, the driver's emergency braking or rapid steering wheel turn may cause abnormal driving data, but it does not mean that the vehicle has definitely been involved in an accident. Therefore, based on the detection results of the target statistical model, the detection results of the deep learning model will be combined for comprehensive analysis to improve the accuracy of vehicle accident detection.

[0084] In practical implementation, since vehicle network sequence data is time series data, deep learning models can use neural networks that are good at time series analysis, such as Long Short-Term Memory (LSTM) networks. As a type of time recurrent neural network, LSTM can effectively process and predict important events with very long intervals and delays in time series.

[0085] In this embodiment, an accident identification is performed on vehicle network sequence data using a pre-trained deep learning model, which enables accurate detection of vehicle driving accidents.

[0086] S104: Based on at least one abnormal data detection result and an accident detection result, determine whether the target vehicle has been involved in a driving accident.

[0087] It should be noted that the abnormal data detection result is the probability value output by the corresponding target statistical model, which represents the probability of the target vehicle being involved in a driving accident; while the accident detection result is the probability value output by the deep learning model, which represents the probability of the target vehicle being involved in a driving accident.

[0088] In practical implementation, weights can be set for at least one abnormal data detection result and one accident detection result according to actual needs. Then, by weighting the abnormal data detection result and the accident detection result, the final target detection result can be obtained. Based on the target detection result, it can be determined whether the target vehicle has been involved in a driving accident.

[0089] In this implementation, on the one hand, the powerful trend analysis capabilities of the ARMA model can be utilized to detect anomalies in various driving data within the vehicle-to-everything (V2X) sequence data; on the other hand, the LSTM model is used for collision event prediction, fully leveraging its advantages in processing time-series data such as V2X sequence data. By combining the target statistical model and the deep learning model, not only can the V2X sequence data be comprehensively analyzed from multiple perspectives, improving the accuracy of vehicle accident detection, but it also facilitates the introduction of more data types of V2X sequence data, enabling full utilization of large-scale V2X data with diverse data types, thus exhibiting stronger scalability and versatility.

[0090] In one feasible implementation, at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result; S102 may specifically include the following sub-steps:

[0091] S102-1: Based on vehicle network sequence data, obtain acceleration sequence data, steering wheel speed sequence data and / or parking behavior sequence data.

[0092] In this embodiment, considering that abnormal acceleration, abnormal steering wheel speed, and abnormal parking behavior usually occur when a vehicle accident occurs, an abnormal acceleration detection model, an abnormal steering wheel speed detection model, and an abnormal parking behavior detection model will be set up to detect anomalies in the above three types of vehicle network sequence data.

[0093] In this embodiment, the vehicle network sequence data is vehicle driving data stored according to a preset format. For example, the vehicle network sequence data can be written to the storage space in a row-based storage manner. When the data interval of the vehicle network sequence data is one second, one row of the current vehicle network sequence data is written every second. Each column represents a different data type, such as the first column representing acceleration, the second column representing steering wheel speed, and the third column representing parking behavior.

[0094] In practice, acceleration sequence data, steering wheel speed sequence data, and parking behavior sequence data can be extracted from each column of vehicle network sequence data.

[0095] S102-2: Input acceleration sequence data into the acceleration anomaly detection model and output the acceleration anomaly detection results.

[0096] In this embodiment, acceleration sequence data refers to the acceleration information generated by the target vehicle according to a time sequence, used to characterize the acceleration variation pattern of the target vehicle. By detecting abnormal accelerations in the acceleration sequence data using an acceleration anomaly detection model, it is possible to determine whether a driving accident has occurred in the target vehicle from the perspective of acceleration. In other words, the acceleration anomaly detection result is used to characterize the probability of a driving accident occurring in the target vehicle's acceleration dimension.

[0097] S102-3: Input the steering wheel speed sequence data into the steering wheel speed anomaly detection model and output the steering wheel speed anomaly detection result.

[0098] In this embodiment, the steering wheel speed sequence data is the steering wheel speed information generated by the target vehicle according to a time sequence, used to characterize the steering wheel speed variation pattern of the target vehicle. By detecting abnormal steering wheel speeds in the steering wheel speed sequence data using a steering wheel speed anomaly detection model, it is possible to determine whether a driving accident has occurred in the target vehicle from the perspective of steering wheel speed. In other words, the steering wheel speed anomaly detection result is used to characterize the probability of a driving accident occurring in the dimension of steering wheel speed of the target vehicle.

[0099] S102-4: Input the parking behavior sequence data into the parking behavior anomaly detection model and output the parking behavior anomaly detection results.

[0100] In this embodiment, parking behavior sequence data refers to the parking behavior information of the target vehicle generated in a time sequence, used to characterize the changing patterns of the target vehicle's parking behavior. By detecting abnormal parking behaviors in the parking behavior sequence data using a parking behavior anomaly detection model, it is possible to determine whether the target vehicle has been involved in a driving accident from the perspective of parking behavior. In other words, the parking behavior anomaly detection result is used to characterize the probability of the target vehicle having a driving accident in the dimension of parking behavior.

[0101] It should be noted that abnormal parking behaviors include: not straightening the steering wheel when exiting the vehicle, not locking the vehicle for an extended period after exiting, and / or turning on the hazard lights after parking. Specifically, if the driver's side door is detected to be open and the steering wheel angle is greater than a preset angle, it is determined that there is abnormal parking behavior of not straightening the steering wheel when exiting the vehicle; if the driver's side door is detected to be open and no locking command is received within a preset time, it is determined that there is abnormal parking behavior of not locking the vehicle for an extended period after exiting; if the vehicle is detected to be parked and the hazard lights are on, it is determined that there is abnormal parking behavior of turning on the hazard lights after parking.

[0102] For example, abnormal parking behavior can be a combination of multiple behaviors. For instance, if the vehicle stops, turns on its hazard lights after parking, and the driver gets out of the vehicle in sequence within a preset time, then the target vehicle is considered to have engaged in abnormal parking behavior.

[0103] In this embodiment, by setting corresponding target statistical models for different types of vehicle network sequence data, comprehensive detection of various abnormal driving data can be achieved. Specifically, a detection model library can be constructed to store multiple pre-trained target statistical models. These models can be added to or removed from the library according to actual detection needs. For example, a vehicle speed anomaly detection model can be added to the library to detect abnormal vehicle speeds in the vehicle network sequence data.

[0104] In one feasible implementation, S103 may specifically include the following sub-steps:

[0105] S103-1: Based on vehicle network sequence data, obtain vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data.

[0106] In this embodiment, the vehicle network sequence data also includes vehicle speed data. By extracting the vehicle speed data from the vehicle network sequence data according to the time sequence, vehicle speed sequence data can also be obtained, so that the deep learning model can make full use of the diverse types of vehicle network data and improve detection accuracy.

[0107] S103-2: Input vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data into the deep learning model to obtain accident detection results.

[0108] In this embodiment, since the deep learning model has fully learned the abnormal features of vehicle speed, acceleration and steering wheel speed in the vehicle network sequence data during the model training phase, by inputting the vehicle speed sequence data, acceleration sequence data and steering wheel speed sequence data into the deep learning model, the deep learning model can extract and identify features of each data and directly output accident detection results.

[0109] In this embodiment, by comprehensively considering sequence data of various data types, the detection accuracy of deep learning models can be effectively improved.

[0110] In one feasible implementation, S104 may specifically include the following sub-steps:

[0111] S104-1: Determine the first detection time for abnormal acceleration detection results, the second detection time for abnormal steering wheel speed detection results, the third detection time for abnormal parking behavior detection results, and the fourth detection time for accident detection results;

[0112] In this embodiment, abnormal driving data are typically temporally correlated. For example, taking vehicle stopping as a dividing point, abnormal acceleration and steering wheel speed usually occur before the vehicle stops, while abnormal stopping behavior usually occurs after the vehicle stops. Therefore, to fully consider the temporal correlation of abnormal driving data of different types at the time of a driving accident, when the target statistical model outputs the anomaly detection results, a corresponding time label will be added to each anomaly detection result. This time label is used to characterize the detection time of the corresponding anomaly detection result; that is, time labels representing the first detection time, the second detection time, and the third detection time will be added to the acceleration anomaly detection result, the steering wheel speed anomaly detection result, and the stopping behavior anomaly detection result, respectively. In addition, when the deep learning model outputs the accident detection results, a time label representing the fourth detection time will also be added to the accident detection results.

[0113] S104-2: If the preset conditions are met at the first detection time, the second detection time, the third detection time, and the fourth detection time, determine whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection results, steering wheel speed anomaly detection results, parking behavior anomaly detection results, and accident detection results.

[0114] In this embodiment, since the accident detection results are for driving accidents, when a driving accident actually occurs, the acceleration anomaly detection results and steering wheel speed anomaly detection results usually appear before the accident detection results, while the parking behavior anomaly detection results usually appear after the accident detection results. Therefore, the preset conditions can be set as follows: the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time. That is, when the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time, it can be determined whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection results, steering wheel speed anomaly detection results, parking behavior anomaly detection results, and accident detection results.

[0115] In this embodiment, by considering the temporal correlation of various detection results, misjudgments can be effectively avoided, thereby further improving detection accuracy.

[0116] In its specific implementation, step S104-2, which determines whether the target vehicle has been involved in a driving accident based on the results of acceleration anomaly detection, steering wheel speed anomaly detection, parking behavior anomaly detection, and accident detection, may include the following sub-steps:

[0117] S104-2-1: The results of acceleration anomaly detection, steering wheel speed anomaly detection, parking behavior anomaly detection, and accident detection are weighted and summed to obtain the accident score.

[0118] In practice, the accident score can be calculated using the following formula:

[0119] T=w1*T1+w2*T2+w3*T3+w4*T4;

[0120] w1+w2+w3+w4=1;

[0121] Where T represents the accident score, T1 represents the acceleration anomaly detection result, w1 represents the first weight value corresponding to the acceleration anomaly detection result, T2 represents the steering wheel speed anomaly detection result, w2 represents the second weight value corresponding to the steering wheel speed anomaly detection result, T3 represents the parking behavior anomaly detection result, w3 represents the third weight value corresponding to the parking behavior anomaly detection result, and T4 represents the accident detection result, w4 represents the accident detection result.

[0122] It should be noted that the first weight value, the second weight value, the third weight value, and the fourth weight value can be set according to actual needs, and this implementation method does not impose specific restrictions on them.

[0123] For example, a higher weight value can be assigned to the accident detection result, and a lower weight value can be assigned to the acceleration anomaly detection result, steering wheel speed anomaly detection result, and parking behavior anomaly detection result. For instance, the fourth weight value can be set to 0.4, and the first, second, and third weight values ​​can be set to 0.2 each.

[0124] S104-2-2: If the accident score is greater than the score threshold, it is determined that the target vehicle has been involved in a driving accident.

[0125] In this embodiment, the scoring threshold is used to characterize the probability that the target vehicle actually causes a driving accident. Specifically, when the detected accident score value is greater than the scoring threshold, it indicates that the target vehicle has a high probability of causing a driving accident, and the target vehicle is considered to have caused a driving accident; while when the accident score value is less than or equal to the scoring threshold, the target vehicle is considered not to have caused a driving accident.

[0126] In one example, the scoring threshold is set to 0.8, the first, second, and third weight values ​​are set to 0.2, and the fourth weight value can be set to 0.4. When the acceleration anomaly detection result, steering wheel speed anomaly detection result, parking behavior anomaly detection result, and accident detection result are 0.86, 0.84, 0.78, and 0.75 respectively, the accident score is calculated as 0.4 * 0.85 + 0.2 * 0.8 + 0.2 * 0.9 + 0.2 * 0.75 = 0.82, which is greater than the scoring threshold, thus determining that the target vehicle has been involved in a driving accident.

[0127] In a feasible implementation, considering that the probability of abnormal driving data and the probability of driving accidents may differ when a vehicle is traveling on different road sections, the vehicle accident detection method may further include the following steps before S104-2-2 to further improve the accuracy of accident detection:

[0128] S201: Obtain road condition information of the road where the target vehicle is located.

[0129] In practice, the current location information of the target vehicle can be obtained through the target vehicle's map navigation system, and the current road segment where the current location is located can be determined. Then, based on a pre-built road condition database, the road condition information corresponding to the current road segment can be determined. The road condition database is used to store road condition information for different road segments.

[0130] In practical implementation, a pre-trained road condition detection model can be used to identify the road condition information of the target vehicle. Specifically, feature data of the road to be detected can be obtained, including vehicle driving feature data and environmental feature data of the road where the target vehicle is located. By inputting the feature data of the road to be detected into the road condition detection model, road condition information can be output. Among them, vehicle driving feature data includes vehicle speed, acceleration, steering wheel speed and / or vehicle attitude; environmental feature data includes congestion information (e.g., smooth traffic, heavy congestion, light congestion, etc.) and weather features (e.g., temperature, rainfall, snowfall, etc.).

[0131] It should be noted that road condition information is used to characterize the difficulty of driving on the road where the target vehicle is located. For example, road condition information can be divided into several levels, with higher levels indicating that the road is more difficult to traverse. For instance, there are three road condition levels: Level 1, Level 2, and Level 3, corresponding to good, average, and severe road conditions, respectively.

[0132] It should be noted that road condition levels can be classified according to actual needs, and this implementation method does not impose specific restrictions.

[0133] S202: Determine the scoring threshold based on road condition information.

[0134] In this implementation, different road condition information corresponds to different scoring thresholds. In a specific implementation, a two-dimensional table representing the mapping relationship between road condition levels and scoring thresholds can be pre-constructed. Then, after determining the corresponding road condition level based on the road condition information, the scoring threshold corresponding to the road condition level can be determined by looking up the table.

[0135] For example, considering that the probability of abnormal driving data such as abnormal acceleration and abnormal steering wheel speed is small when the road conditions are good, the scoring threshold can be set as the first scoring threshold when the road conditions are good. However, when the road conditions are bad, abnormal driving data such as abnormal acceleration may occur when the driver is operating the vehicle normally. In this case, in order to avoid misjudgment, the scoring threshold can be set as the second scoring threshold, where the second scoring threshold is greater than the first scoring threshold.

[0136] In this embodiment, by dynamically setting the scoring threshold based on the actual road conditions of the target vehicle, the misjudgment of accidents caused by static thresholds can be effectively avoided, thereby further improving the accuracy of accident detection.

[0137] In a feasible implementation, at least one objective statistical model and a deep learning model are trained in the following manner:

[0138] S301: For any target statistical model, based on the target data type of the abnormal driving data used by the target statistical model for detection, extract the target data sequence of the target data type from the historical vehicle network sequence data.

[0139] It should be noted that the historical vehicle network sequence data is obtained by collecting driving data from vehicles with the same model as the target vehicle.

[0140] In this embodiment, historical vehicle network sequence data is extracted according to the target data type. The extracted target data sequence may include, but is not limited to, historical vehicle speed sequence data, historical acceleration sequence data, historical steering wheel speed sequence data, and historical parking behavior sequence data.

[0141] For example, to train an acceleration anomaly detection model, historical acceleration sequence data corresponding to the acceleration data type can be extracted from historical vehicle network sequence data; to train a steering wheel speed anomaly detection model, historical steering wheel speed sequence data corresponding to the steering wheel speed data type can be extracted from historical vehicle network sequence data; to train a parking behavior anomaly detection model, historical parking behavior sequence data corresponding to the parking behavior data type can be extracted from historical vehicle network sequence data.

[0142] In this embodiment, in order to improve the model training effect, after acquiring historical vehicle network sequence data and extracting the target data sequence, the historical vehicle network sequence data and the target data sequence will also be processed by missing value filling and normalization.

[0143] S302: Label the target data sequence and historical vehicle network sequence data with accident tags to obtain the first model training samples and the second model training samples, respectively.

[0144] In this embodiment, for any target data sequence, it can be labeled by historical accident data to obtain the accident label corresponding to the target data sequence, so as to obtain the first model training sample.

[0145] In this embodiment, for historical vehicle network sequence data, the historical vehicle network sequence data is treated as a whole and labeled with accident tags to obtain the second model training samples.

[0146] It should be noted that the incident label includes whether an incident occurred and the time of the incident. In the specific implementation, if the incident label indicates that an incident has occurred, the incident label can be represented by 1; if the incident label indicates that no incident has occurred, the incident label can be represented by 0.

[0147] S303: Using the training samples of the first model, train the initial statistical model to obtain the target statistical model.

[0148] It should be noted that when there are multiple target data sequences, the same number of first model training samples will be obtained. First model training samples of different data types are used to train different initial statistical models. For example, when the data type of the first model training samples is acceleration, the initial statistical model trained with the first model training samples can obtain an acceleration anomaly detection model; when the data type of the first model training samples is steering wheel speed, the initial statistical model trained with the first model training samples can obtain a steering wheel speed anomaly detection model.

[0149] In the specific implementation, based on the accident labels, the training samples of the first model can be divided into positive samples and negative samples. Positive samples are sample sequences composed of historical driving accident data combined in time sequence, meaning they consist of samples with an accident label of 1. Negative samples are sample sequences composed of historical normal driving data combined in time sequence, meaning they consist of samples with an accident label of 0. Using these positive and negative samples, the initial statistical model is trained through binary classification to obtain the target statistical model.

[0150] S304: Train the initial deep learning model using the training samples of the second model to obtain the deep learning model.

[0151] In this embodiment, the training samples of the second model are a sequence of samples with an accident label of 1 in the historical vehicle network sequence data. By outputting the training samples of the second model into the initial deep learning model, the initial deep learning model learns the accident features and outputs the driving accident prediction value corresponding to the sample sequence. Then, backpropagation is performed based on the difference between the driving accident prediction value and the accident label to update the network parameters of the initial deep learning model until the initial deep learning model reaches the training cutoff condition, thus obtaining the deep learning model.

[0152] Secondly, based on the same inventive concept, and referring to... Figure 2 This application provides a vehicle accident detection device 200, which includes:

[0153] The data acquisition module 201 is used to acquire the vehicle network sequence data of the target vehicle;

[0154] The first detection module 202 is used to input vehicle network sequence data into at least one target statistical model to obtain at least one abnormal data detection result; different target statistical models are used to detect abnormal driving data of different types.

[0155] The second detection module 203 is used to input vehicle network sequence data into a deep learning model to obtain accident detection results;

[0156] The accident determination module 204 is used to determine whether the target vehicle has been involved in a driving accident based on at least one abnormal data detection result and an accident detection result.

[0157] In one embodiment of this application, at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result; the first detection module 202 includes:

[0158] The first data extraction submodule is used to obtain acceleration sequence data, steering wheel speed sequence data and / or parking behavior sequence data based on vehicle network sequence data;

[0159] The acceleration detection submodule is used to input acceleration sequence data into the acceleration anomaly detection model and output the acceleration anomaly detection result.

[0160] The steering wheel speed detection submodule is used to input the steering wheel speed sequence data into the steering wheel speed anomaly detection model and output the steering wheel speed anomaly detection result.

[0161] The parking behavior detection submodule is used to input parking behavior sequence data into the parking behavior anomaly detection model and output the parking behavior anomaly detection results.

[0162] In one embodiment of this application, the second detection module 203 includes:

[0163] The second data extraction submodule is used to obtain vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data based on vehicle network sequence data;

[0164] The accident detection submodule is used to input vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data into a deep learning model to obtain accident detection results.

[0165] In one embodiment of this application, the accident determination module 204 includes:

[0166] The detection time determination submodule is used to determine the first detection time for acceleration anomaly detection results, the second detection time for steering wheel speed anomaly detection results, the third detection time for parking behavior anomaly detection results, and the fourth detection time for accident detection results.

[0167] The accident determination submodule is used to determine whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, steering wheel speed anomaly detection result, parking behavior anomaly detection result, and accident detection result, provided that preset conditions are met at the first detection time, second detection time, third detection time, and fourth detection time.

[0168] In one embodiment of this application, the vehicle accident detection device 200 further includes:

[0169] The condition determination module is used to determine whether the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions when the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time.

[0170] In one embodiment of this application, the accident determination submodule includes:

[0171] The weighted summation unit is used to perform weighted summation on the acceleration anomaly detection results, steering wheel speed anomaly detection results, parking behavior anomaly detection results, and accident detection results to obtain an accident score.

[0172] The accident determination unit is used to determine that a target vehicle has been involved in a driving accident when the accident score is greater than the score threshold.

[0173] In one embodiment of this application, the vehicle accident detection device further includes a model training module, which includes:

[0174] The third data extraction submodule is used to extract the target data sequence of the target data type from historical vehicle network sequence data, based on the target data type of the abnormal driving data detected by the target statistical model for any target statistical model.

[0175] The data labeling submodule is used to label the target data sequence and historical vehicle network sequence data with accident labels, and obtain the first model training samples and the second model training samples, respectively; the accident label includes whether an accident has occurred and the time of the accident;

[0176] The first model training submodule is used to train the initial statistical model using the first model training samples to obtain the target statistical model;

[0177] The second model training submodule is used to train the initial deep learning model using the second model training samples to obtain the deep learning model.

[0178] It should be noted that the specific implementation of the vehicle accident detection device 200 in this application embodiment refers to the specific implementation of the vehicle accident detection method proposed in the first aspect of the above-mentioned application embodiment, and will not be repeated here.

[0179] Thirdly, based on the same inventive concept, embodiments of this application provide a storage medium storing machine-executable instructions, which, when executed by a processor, implement the vehicle accident detection method proposed in the first aspect of this application.

[0180] It should be noted that the specific implementation of the storage medium in the embodiments of this application refers to the specific implementation of the vehicle accident detection method proposed in the first aspect of this application, and will not be repeated here.

[0181] Thirdly, based on the same inventive concept, embodiments of this application provide a vehicle including a vehicle accident detection system, which is used to execute the vehicle accident detection method proposed in the first aspect of this application.

[0182] It should be noted that the specific implementation method of the vehicle in this application embodiment refers to the specific implementation method of the vehicle accident detection method proposed in the first aspect of this application, and will not be repeated here.

[0183] Fifthly, based on the same inventive concept, referring to Figure 3 This application provides a monitoring terminal 300, including a processor 301 and a memory 302; the memory 302 stores machine-executable instructions that can be executed by the processor 301, and the processor 301 is used to execute the machine-executable instructions to implement the vehicle accident detection method proposed in the first aspect of this application.

[0184] It should be noted that the specific implementation of the monitoring terminal 300 in this application embodiment refers to the specific implementation of the vehicle accident detection method proposed in the first aspect of this application, and will not be repeated here.

[0185] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0186] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0187] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0188] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0189] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0190] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0191] The present invention provides a detailed description of a vehicle accident detection method, device, storage medium, vehicle, and monitoring terminal. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for detecting vehicle accidents, characterized in that, The method includes: Obtain the vehicle-to-everything (V2X) sequence data of the target vehicle; The vehicle network sequence data is input into at least one pre-trained target statistical model to obtain at least one abnormal data detection result; different target statistical models are used to detect abnormal driving data of different types. The vehicle network sequence data is input into a pre-trained deep learning model to obtain accident detection results; Based on the at least one abnormal data detection result and the accident detection result, it is determined whether the target vehicle has been involved in a driving accident; Wherein, the at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; the at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result; The step of inputting the vehicle network sequence data into at least one target statistical model to obtain at least one abnormal data detection result includes: Based on the vehicle network sequence data, acceleration sequence data, steering wheel speed sequence data, and / or parking behavior sequence data are obtained; The acceleration sequence data is input into the acceleration anomaly detection model, and the acceleration anomaly detection result is output; and / or, The steering wheel speed sequence data is input into the steering wheel speed anomaly detection model, and the steering wheel speed anomaly detection result is output; and / or, The parking behavior sequence data is input into the parking behavior anomaly detection model, and the parking behavior anomaly detection result is output. The step of determining whether the target vehicle has been involved in a driving accident based on the at least one abnormal data detection result and the accident detection result includes: The first detection time for the acceleration anomaly detection result, the second detection time for the steering wheel speed anomaly detection result, the third detection time for the parking behavior anomaly detection result, and the fourth detection time for the accident detection result are determined. If the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions, based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, it is determined whether the target vehicle has been involved in a driving accident.

2. The vehicle accident detection method according to claim 1, characterized in that, The steps of inputting the vehicle network sequence data into a deep learning model to obtain accident detection results include: Based on the vehicle network sequence data, vehicle speed sequence data, acceleration sequence data, and steering wheel speed sequence data are obtained; The vehicle speed sequence data, the acceleration sequence data, and the steering wheel rotation speed sequence data are input into the deep learning model to obtain the accident detection results.

3. The vehicle accident detection method according to claim 1, characterized in that, Before the step of determining whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, provided that the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions, the method further includes: If the first detection time and the second detection time are earlier than the fourth detection time and the fourth detection time is earlier than the third detection time, then the first detection time, the second detection time, the third detection time and the fourth detection time are determined to satisfy the preset condition.

4. The vehicle accident detection method according to claim 1, characterized in that, The steps for determining whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection results, the steering wheel speed anomaly detection results, the parking behavior anomaly detection results, and the accident detection results include: The accident score is obtained by weighted summation of the acceleration anomaly detection results, the steering wheel speed anomaly detection results, the parking behavior anomaly detection results, and the accident detection results. If the accident score is greater than the score threshold, it is determined that the target vehicle has been involved in a driving accident.

5. The vehicle accident detection method according to claim 1, characterized in that, The at least one target statistical model and the deep learning model are trained in the following manner: For any of the target statistical models, based on the target data type of the abnormal driving data detected by the target statistical model, the target data sequence of the target data type is extracted from the historical vehicle network sequence data; Accident tags are labeled on the target data sequence and the historical vehicle network sequence data to obtain the first model training samples and the second model training samples, respectively; the accident tags include whether an accident has occurred and the time of the accident. The initial statistical model is trained using the training samples of the first model to obtain the target statistical model; The initial deep learning model is trained using the training samples of the second model to obtain the deep learning model.

6. A vehicle accident detection device, characterized in that, The device includes: The data acquisition module is used to acquire the vehicle-to-everything (V2X) sequence data of the target vehicle. The first detection module is used to input the vehicle network sequence data into at least one target statistical model to obtain at least one abnormal data detection result; different target statistical models are used to detect abnormal driving data of different types. The second detection module is used to input the vehicle network sequence data into a deep learning model to obtain accident detection results; An accident determination module is used to determine whether the target vehicle has been involved in a driving accident based on the at least one abnormal data detection result and the accident detection result; Wherein, the at least one target statistical model includes an acceleration anomaly detection model, a steering wheel speed anomaly detection model, and / or a parking behavior anomaly detection model; the at least one abnormal data detection result includes an acceleration anomaly detection result, a steering wheel speed anomaly detection result, and / or a parking behavior anomaly detection result; the first detection module includes: The first data extraction submodule is used to obtain acceleration sequence data, steering wheel speed sequence data and / or parking behavior sequence data based on the vehicle network sequence data; An acceleration detection submodule is used to input the acceleration sequence data into the acceleration anomaly detection model and output the acceleration anomaly detection result. The steering wheel speed detection submodule is used to input the steering wheel speed sequence data into the steering wheel speed anomaly detection model and output the steering wheel speed anomaly detection result. The parking behavior detection submodule is used to input the parking behavior sequence data into the parking behavior anomaly detection model and output the parking behavior anomaly detection result. The accident determination module includes: The detection time determination submodule is used to determine the first detection time of the acceleration anomaly detection result, the second detection time of the steering wheel speed anomaly detection result, the third detection time of the parking behavior anomaly detection result, and the fourth detection time of the accident detection result; The accident determination submodule is used to determine whether the target vehicle has been involved in a driving accident based on the acceleration anomaly detection result, the steering wheel speed anomaly detection result, the parking behavior anomaly detection result, and the accident detection result, provided that the first detection time, the second detection time, the third detection time, and the fourth detection time meet preset conditions.

7. A storage medium, characterized in that, The storage medium stores machine-executable instructions, which, when executed by a processor, implement the vehicle accident detection method as described in any one of claims 1-5.

8. A vehicle, characterized in that, The system includes a vehicle accident detection system, which is used to perform the vehicle accident detection method as described in any one of claims 1-5.

9. A monitoring terminal, characterized in that, The device includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the vehicle accident detection method as described in any one of claims 1-5.

Images