Cloud collaborative vehicle collision identification and early warning method, device and equipment and storage medium

Abstract

The application discloses a kind of cloud collaborative vehicle collision identification early warning method, device, equipment and storage medium, the method is trained by cloud first collision identification model adapted to vehicle end hardware resource, and the first collision identification model is sent to the vehicle end by communication network;The first collision identification model is monitored to the real-time acquisition vehicle operation data by the vehicle end operation, when identifying potential collision event, obtain the target vehicle data containing the preset time period before and after the time when the potential collision event occurs, and the target vehicle data is sent to the cloud end;The target vehicle data is received by the cloud end, and the second collision identification model is used to analyze the target vehicle data twice to obtain collision analysis result, and generate early warning information according to the collision analysis result;It can effectively consider the real-time and accuracy of collision identification, while reducing communication cost and ensuring the integrity of key data.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology for automobiles, and in particular to a cloud-based collaborative vehicle collision recognition and early warning method, device, equipment, and storage medium. Background Technology

[0002] With the development of intelligent connected vehicles, collision detection has become a core competency for OEMs. Relevant policies and market demands are driving its upgrade, but existing technologies have obvious pain points.

[0003] Mainstream collision detection solutions either suffer from an imbalance between accuracy and real-time performance, or have low data transmission efficiency, lack a closed loop for end-to-cloud collaboration, and are disconnected from after-sales rescue coordination.

[0004] T-BOX has the capability to collect collision-related signals, providing possibilities for optimization solutions.

[0005] The existing technology CN116279500B - a vehicle collision recognition method, uses vehicle network big data and machine learning models, combined with vehicle status and user behavior, to achieve multi-dimensional recognition of vehicle collisions. It solves the problems of low recognition accuracy and high cost in the existing technology, and provides a low-cost, high-accuracy recognition method and after-sales care service.

[0006] However, existing technologies have the following drawbacks: 1. Single-vehicle-side rule judgment, weak generalization ability for complex working conditions; 2. There is no lightweight vehicle-side small-scale OTA mechanism, resulting in extremely high iteration costs. Summary of the Invention

[0007] The main objective of this invention is to provide a cloud-based collaborative vehicle collision recognition and early warning method, device, equipment, and storage medium, aiming to solve the technical problems in existing vehicle collision recognition schemes that are difficult to balance real-time performance and recognition accuracy, have difficulty in iteratively updating vehicle-side models, and are prone to losing key data before and after a collision.

[0008] In a first aspect, the present invention provides a cloud-based collaborative vehicle collision recognition and early warning method, the cloud-based collaborative vehicle collision recognition and early warning method comprising the following steps: A first collision recognition model adapted to vehicle-side hardware resources is trained in the cloud and then sent to the vehicle-side via a communication network. The vehicle terminal runs the first collision recognition model to monitor the real-time collected vehicle operation data. When a potential collision event is identified, the target vehicle data within a preset time period before and after the time of the potential collision event is obtained and the target vehicle data is sent to the cloud. The system receives the target vehicle data via the cloud, performs secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generates warning information based on the collision analysis results.

[0009] Optionally, the step of training a first collision recognition model adapted to the vehicle's hardware resources via the cloud and sending the first collision recognition model to the vehicle via a communication network includes: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, a first collision recognition model adapted to vehicle-side hardware resources is trained based on the historical operational data and the accident case data. The first collision recognition model is sent to the vehicle via over-the-air download through a communication network.

[0010] Optionally, the step of aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, and training a first collision recognition model adapted to vehicle-side hardware resources based on the historical operational data and the accident case data, includes: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, an initial collision recognition model is trained using deep learning algorithms based on the historical operational data and the accident case data. The computing power capacity and storage space limitations of the vehicle-mounted remote control terminal are obtained. The initial collision recognition model is compressed and optimized through model pruning and parameter quantization to generate the first collision recognition model.

[0011] Optionally, sending the first collision recognition model to the vehicle via an over-the-air download through a communication network includes: The first collision recognition model is encapsulated into a remote upgrade package with encrypted signature processing via the cloud; The remote upgrade package is pushed to the vehicle via an over-the-air download through a communication network.

[0012] Optionally, the step of monitoring real-time vehicle operation data by running the first collision recognition model on the vehicle end, and when a potential collision event is identified, acquiring target vehicle data within a preset time period before and after the time of the potential collision event, and sending the target vehicle data to the cloud, includes: The vehicle-mounted remote information control terminal loads the first collision recognition model in the local runtime environment, monitors the real-time collected vehicle operation data, obtains real-time feature vectors, and determines whether there is a potential collision event based on the real-time feature vectors. When a potential collision event is detected, target vehicle data within a preset time period before and after the time of the potential collision event is obtained, and the target vehicle data is sent to the cloud.

[0013] Optionally, the step of loading the first collision recognition model in the local runtime environment through the vehicle-mounted remote information control terminal, monitoring the real-time collected vehicle operation data, obtaining real-time feature vectors, and determining whether a potential collision event exists based on the real-time feature vectors includes: The first collision recognition model is loaded in the local runtime environment through the vehicle-mounted remote information control terminal, and the raw sensor data is periodically read from the vehicle controller local area network bus. Real-time feature vectors are extracted from the raw sensor data through a sliding window mechanism. The real-time feature vector is input into the first collision recognition model for inference calculation. When the collision probability value output by the first collision recognition model exceeds a preset probability threshold, a potential collision event is identified.

[0014] Optionally, when a potential collision event is detected, acquiring target vehicle data within a preset time period before and after the occurrence of the potential collision event, and sending the target vehicle data to the cloud, includes: When the potential collision event is detected, the vehicle terminal calls the circular buffer maintained in memory to extract a continuous data segment centered on the trigger time, extending forward for a first preset time and backward for a second preset time as the target vehicle data. After the target vehicle data is formatted, compressed, and encrypted, a communication connection with the cloud is established through a mobile communication module. The processed target vehicle data is uploaded to the cloud.

[0015] Optionally, the target vehicle data includes at least one of the following: raw airbag controller signal, wheel speed signal, vehicle stability system status signal, and GPS location information.

[0016] Optionally, uploading the processed target vehicle data to the cloud includes: The current network status is monitored during the process of uploading the processed target vehicle data to the cloud. When an upload interruption is detected, the current breakpoint position is recorded, and the upload continues from the current breakpoint position after the network is restored; When an upload failure is detected, a preset number of retries will be performed until the upload is successful or the maximum number of retries is reached.

[0017] Optionally, the step of receiving the target vehicle data through the cloud, performing secondary analysis on the target vehicle data using a second collision recognition model to obtain a collision analysis result, and generating warning information based on the collision analysis result includes: The target vehicle data is received through the secure communication interface of the cloud, and the target vehicle data is decrypted, decompressed, and its integrity is verified. The second collision recognition model is used to perform a second in-depth analysis on the verified target vehicle data to obtain the collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model and is used to identify the collision severity level, collision type, damaged parts and occupant injury risk. Based on the collision analysis results, corresponding early warning information is generated and fed back.

[0018] Optionally, generating and feeding back corresponding early warning information based on the collision analysis results includes: Based on the collision analysis results, a visualized early warning message is generated, which includes vehicle location information, accident time, collision level, and suggested rescue measures. The warning information is pushed to the after-sales service platform, emergency rescue call center or user terminal in real time through the application programming interface.

[0019] Optionally, after receiving the target vehicle data via the cloud, performing secondary analysis on the target vehicle data using the second collision recognition model to obtain collision analysis results, and generating warning information based on the collision analysis results, the cloud-based collaborative vehicle collision recognition and warning method further includes: Obtain accident data from the collision analysis results that have been manually or automatically confirmed, and add the accident data as new labeled samples to the historical training dataset; The first collision recognition model is retrained and optimized based on the updated historical training dataset to generate an updated first collision recognition model. The updated first collision recognition model is sent to the vehicle via a communication network.

[0020] Secondly, to achieve the above objectives, the present invention also proposes a cloud-based collaborative vehicle collision recognition and early warning device, the cloud-based collaborative vehicle collision recognition and early warning device comprising: The model adaptation module is used to train a first collision recognition model adapted to the vehicle's hardware resources via the cloud, and send the first collision recognition model to the vehicle via a communication network. The data acquisition module is used to monitor the real-time vehicle operation data by running the first collision recognition model on the vehicle end. When a potential collision event is identified, the module acquires target vehicle data within a preset time period before and after the time of the potential collision event and sends the target vehicle data to the cloud. The analysis and early warning module is used to receive the target vehicle data through the cloud, perform secondary analysis on the target vehicle data using the second collision recognition model to obtain collision analysis results, and generate early warning information based on the collision analysis results.

[0021] Thirdly, to achieve the above objectives, the present invention also proposes a cloud-based collaborative vehicle collision recognition and warning device, which includes: a memory, a processor, and a cloud-based collaborative vehicle collision recognition and warning program stored in the memory and executable on the processor. The cloud-based collaborative vehicle collision recognition and warning program is configured to implement the steps of the cloud-based collaborative vehicle collision recognition and warning method described above.

[0022] Fourthly, to achieve the above objectives, the present invention also proposes a storage medium storing a cloud-based collaborative vehicle collision recognition and warning program, wherein the cloud-based collaborative vehicle collision recognition and warning program, when executed by a processor, implements the steps of the cloud-based collaborative vehicle collision recognition and warning method described above.

[0023] The cloud-based collaborative vehicle collision recognition and early warning method proposed in this invention trains a first collision recognition model adapted to the vehicle's hardware resources in the cloud and sends the first collision recognition model to the vehicle via a communication network. The vehicle monitors real-time vehicle operation data by running the first collision recognition model. When a potential collision event is detected, target vehicle data within a preset time period before and after the potential collision event is acquired and sent to the cloud. The cloud receives the target vehicle data, performs secondary analysis using a second collision recognition model to obtain collision analysis results, and generates early warning information based on these results. This method effectively balances the real-time performance and accuracy of collision recognition while reducing communication costs and ensuring the integrity of key data, avoiding the data loss caused by uploading all data. It also ensures the complete capture of key contextual data before and after the collision, significantly reducing the false alarm rate and generating accurate early warning information to support subsequent rescue services, thus achieving a closed-loop technology from recognition to service. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2This is a flowchart illustrating the first embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention; Figure 3 This is a flowchart illustrating the second embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention; Figure 4 This is a flowchart illustrating the third embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention. Figure 5 This is a flowchart illustrating the fourth embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention; Figure 6 This is a flowchart illustrating the fifth embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention; Figure 7 This is a block diagram of the overall system architecture corresponding to the cloud-based collaborative vehicle collision recognition and early warning method of the present invention; Figure 8 This is a functional block diagram of the first embodiment of the cloud-based collaborative vehicle collision recognition and early warning device of the present invention.

[0025] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0026] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0027] The solution of this invention mainly involves: training a first collision recognition model adapted to vehicle-side hardware resources in the cloud, and sending the first collision recognition model to the vehicle via a communication network; monitoring real-time vehicle operation data by running the first collision recognition model on the vehicle; when a potential collision event is detected, acquiring target vehicle data within a preset time period before and after the potential collision event, and sending the target vehicle data to the cloud; receiving the target vehicle data through the cloud, performing secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generating warning information based on the collision analysis results; effectively balancing the real-time performance and accuracy of collision recognition, while reducing communication costs and ensuring the integrity of key data, avoiding the waste of traffic caused by uploading all data, and ensuring the complete capture of key context data before and after the collision, significantly reducing the false alarm rate, thereby generating accurate warning information to support subsequent rescue services, realizing a technical closed loop from recognition to service, and solving the technical problems in existing vehicle collision recognition schemes that are difficult to balance real-time performance and recognition accuracy, difficult to iterate and update vehicle-side models, and prone to loss of key data before and after collisions.

[0028] Reference Figure 1 , Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0029] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU; a communication bus 1002; a user interface 1003; a network interface 1004; and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0030] Those skilled in the art will understand that Figure 1 The device structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0031] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a cloud-based collaborative vehicle collision recognition and warning program.

[0032] The device of this invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and performs the following operations: A first collision recognition model adapted to vehicle-side hardware resources is trained in the cloud and then sent to the vehicle-side via a communication network. The vehicle terminal runs the first collision recognition model to monitor the real-time collected vehicle operation data. When a potential collision event is identified, the target vehicle data within a preset time period before and after the time of the potential collision event is obtained and the target vehicle data is sent to the cloud. The system receives the target vehicle data via the cloud, performs secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generates warning information based on the collision analysis results.

[0033] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, a first collision recognition model adapted to vehicle-side hardware resources is trained based on the historical operational data and the accident case data. The first collision recognition model is sent to the vehicle via over-the-air download through a communication network.

[0034] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, an initial collision recognition model is trained using deep learning algorithms based on the historical operational data and the accident case data. The computing power capacity and storage space limitations of the vehicle-mounted remote control terminal are obtained. The initial collision recognition model is compressed and optimized through model pruning and parameter quantization to generate the first collision recognition model.

[0035] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The first collision recognition model is encapsulated into a remote upgrade package with encrypted signature processing via the cloud; The remote upgrade package is pushed to the vehicle via an over-the-air download through a communication network.

[0036] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The vehicle-mounted remote information control terminal loads the first collision recognition model in the local runtime environment, monitors the real-time collected vehicle operation data, obtains real-time feature vectors, and determines whether there is a potential collision event based on the real-time feature vectors. When a potential collision event is detected, target vehicle data within a preset time period before and after the time of the potential collision event is obtained, and the target vehicle data is sent to the cloud.

[0037] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The first collision recognition model is loaded in the local runtime environment through the vehicle-mounted remote information control terminal, and the raw sensor data is periodically read from the vehicle controller local area network bus. Real-time feature vectors are extracted from the raw sensor data through a sliding window mechanism. The real-time feature vector is input into the first collision recognition model for inference calculation. When the collision probability value output by the first collision recognition model exceeds a preset probability threshold, a potential collision event is identified.

[0038] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: When the potential collision event is detected, the vehicle terminal calls the circular buffer maintained in memory to extract a continuous data segment centered on the trigger time, extending forward for a first preset time and backward for a second preset time as the target vehicle data. After the target vehicle data is formatted, compressed, and encrypted, a communication connection with the cloud is established through a mobile communication module. The processed target vehicle data is uploaded to the cloud.

[0039] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The target vehicle data includes at least one of the following: raw airbag controller signal, wheel speed signal, vehicle stability system status signal, and GPS location information.

[0040] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The current network status is monitored during the process of uploading the processed target vehicle data to the cloud. When an upload interruption is detected, the current breakpoint position is recorded, and the upload continues from the current breakpoint position after the network is restored; When an upload failure is detected, a preset number of retries will be performed until the upload is successful or the maximum number of retries is reached.

[0041] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: The target vehicle data is received through the secure communication interface of the cloud, and the target vehicle data is decrypted, decompressed, and its integrity is verified. The second collision recognition model is used to perform a second in-depth analysis on the verified target vehicle data to obtain the collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model and is used to identify the collision severity level, collision type, damaged parts and occupant injury risk. Based on the collision analysis results, corresponding early warning information is generated and fed back.

[0042] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: Based on the collision analysis results, a visualized early warning message is generated, which includes vehicle location information, accident time, collision level, and suggested rescue measures. The warning information is pushed to the after-sales service platform, emergency rescue call center or user terminal in real time through the application programming interface.

[0043] The device of the present invention calls the cloud-based collaborative vehicle collision recognition and warning program stored in the memory 1005 through the processor 1001, and also performs the following operations: Obtain accident data from the collision analysis results that have been manually or automatically confirmed, and add the accident data as new labeled samples to the historical training dataset; The first collision recognition model is retrained and optimized based on the updated historical training dataset to generate an updated first collision recognition model. The updated first collision recognition model is sent to the vehicle via a communication network.

[0044] This embodiment employs the above-described scheme, training a first collision recognition model adapted to vehicle-side hardware resources in the cloud and transmitting this model to the vehicle via a communication network. The vehicle then runs the first collision recognition model to monitor real-time vehicle operation data. When a potential collision event is detected, target vehicle data within a preset time period before and after the potential collision event is acquired and transmitted to the cloud. The cloud receives the target vehicle data, performs secondary analysis using a second collision recognition model to obtain collision analysis results, and generates warning information based on these results. This approach effectively balances the real-time performance and accuracy of collision recognition while reducing communication costs and ensuring the integrity of key data. It avoids the data waste caused by uploading all data and ensures the complete capture of key contextual data before and after a collision, significantly reducing the false alarm rate. This generates accurate warning information to support subsequent rescue services, achieving a closed-loop technology from recognition to service.

[0045] Based on the above hardware structure, an embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention is proposed.

[0046] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the cloud-based collaborative vehicle collision recognition and early warning method of the present invention.

[0047] In the first embodiment, the cloud-based collaborative vehicle collision recognition and early warning method includes the following steps: Step S10: Train a first collision recognition model adapted to the vehicle's hardware resources through the cloud, and send the first collision recognition model to the vehicle through a communication network.

[0048] It should be noted that by training a first collision recognition model adapted to the vehicle's hardware resources in the cloud and sending the first collision recognition model to the vehicle via a communication network, the advantages of cloud computing power can be used to complete model training and optimization to match the vehicle's limited computing power and storage resources, thereby realizing the remote deployment and dynamic updating of collision recognition capabilities.

[0049] Step S20: The first collision recognition model is run on the vehicle to monitor the real-time vehicle operation data. When a potential collision event is identified, the target vehicle data within a preset time period before and after the time of the potential collision event is obtained, and the target vehicle data is sent to the cloud.

[0050] It should be understood that by running the first collision recognition model on the vehicle end to monitor the real-time collected vehicle operation data, when a potential collision event is identified, the target vehicle data within a preset time period before and after the time of the potential collision event is obtained, and the target vehicle data is sent to the cloud. This can ensure real-time response to collision events while accurately capturing key contextual data before and after the collision using a trigger mechanism.

[0051] Step S30: Receive the target vehicle data through the cloud, perform secondary analysis on the target vehicle data using the second collision recognition model to obtain collision analysis results, and generate warning information based on the collision analysis results.

[0052] It is understood that by receiving the target vehicle data through the cloud, the second collision recognition model can be used to perform secondary analysis on the target vehicle data to obtain collision analysis results, and warning information can be generated based on the collision analysis results. The high-performance computing power of the cloud can be used to perform in-depth verification and multi-dimensional evaluation of suspected collision events reported by the vehicle, effectively reducing the false alarm rate and obtaining detailed accident level and damage information.

[0053] This embodiment employs the above-described scheme, training a first collision recognition model adapted to vehicle-side hardware resources in the cloud and transmitting this model to the vehicle via a communication network. The vehicle then runs the first collision recognition model to monitor real-time vehicle operation data. When a potential collision event is detected, target vehicle data within a preset time period before and after the potential collision event is acquired and transmitted to the cloud. The cloud receives the target vehicle data, performs secondary analysis using a second collision recognition model to obtain collision analysis results, and generates warning information based on these results. This approach effectively balances the real-time performance and accuracy of collision recognition while reducing communication costs and ensuring the integrity of key data. It avoids the data waste caused by uploading all data and ensures the complete capture of key contextual data before and after a collision, significantly reducing the false alarm rate. This generates accurate warning information to support subsequent rescue services, achieving a closed-loop technology from recognition to service.

[0054] Furthermore, Figure 3 This is a flowchart illustrating the second embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention, as shown below. Figure 3 As shown, based on the first embodiment, a second embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention is proposed. In this embodiment, step S10 specifically includes the following steps: Step S11: Gather historical operation data of the entire fleet and confirmed accident case data through the cloud, and train a first collision recognition model adapted to the hardware resources of the vehicle based on the historical operation data and the accident case data.

[0055] It should be noted that by aggregating the historical operation data of the entire fleet and confirmed accident case data in the cloud, and training a first collision recognition model adapted to the hardware resources of the vehicle based on the historical operation data and the accident case data, the generalization ability and recognition accuracy of the model can be improved by utilizing massive and diverse data. At the same time, it ensures that the trained model meets the computing power and storage limitations of the vehicle, and realizes the effective deployment of high-precision models in resource-constrained environments.

[0056] Furthermore, step S11 specifically includes the following steps: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, an initial collision recognition model is trained using deep learning algorithms based on the historical operational data and the accident case data. The computing power capacity and storage space limitations of the vehicle-mounted remote control terminal are obtained. The initial collision recognition model is compressed and optimized through model pruning and parameter quantization to generate the first collision recognition model.

[0057] It should be understood that by aggregating historical operational data of the entire fleet and confirmed accident case data in the cloud, and using deep learning algorithms to train an initial collision recognition model based on the historical operational data and the accident case data, and by obtaining the computing power capacity and storage space limitations of the vehicle-side remote control terminal, the initial collision recognition model is compressed and optimized through model pruning and parameter quantization to generate a first collision recognition model. This approach can utilize massive and diverse data to ensure the basic recognition accuracy and generalization ability of the model, while compression and optimization techniques enable the model to adapt to the limited computing power and storage resources on the vehicle side, ensuring the feasible deployment and efficient operation of the high-precision model on the vehicle-side hardware.

[0058] Step S12: Send the first collision recognition model to the vehicle via over-the-air download through the communication network.

[0059] It is understandable that sending the first collision recognition model to the vehicle via over-the-air download through a communication network enables remote wireless transmission and dynamic deployment of the model file, avoiding the cumbersome process and high cost of physical connection upgrades, and ensuring that the vehicle always has the latest recognition capabilities adapted to the current working conditions.

[0060] Furthermore, step S12 specifically includes the following steps: The first collision recognition model is encapsulated into a remote upgrade package with encrypted signature processing via the cloud; The remote upgrade package is pushed to the vehicle via an over-the-air download through a communication network.

[0061] It should be understood that by encapsulating the first collision recognition model into a remote upgrade package with encrypted signature processing through the cloud, and pushing the remote upgrade package to the vehicle via over-the-air download through the communication network, the security and integrity of the model file during transmission can be ensured, preventing data from being maliciously tampered with or illegally loaded. At the same time, wireless communication technology is used to achieve seamless remote deployment, avoiding the cumbersome operation of physical connection upgrades, and ensuring the reliability and efficiency of the model update process.

[0062] This embodiment, through the above-described scheme, aggregates historical operational data and confirmed accident case data of the entire fleet in the cloud, and trains a first collision recognition model adapted to the hardware resources of the vehicle based on the historical operational data and the accident case data; the first collision recognition model is sent to the vehicle via over-the-air download through a communication network; it can leverage the advantages of the full amount of data aggregated in the cloud to improve the generalization ability and recognition accuracy of the model, while ensuring that the model is adapted to the limited hardware resources of the vehicle to achieve efficient local inference, and realize remote wireless iterative updates of the model through over-the-air download, continuously optimizing the collision recognition capability of the vehicle without physical connection, and reducing operation and maintenance costs.

[0063] Furthermore, Figure 4 This is a flowchart illustrating the third embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention, as shown below. Figure 4 As shown, based on the first embodiment, a third embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention is proposed. In this embodiment, step S20 specifically includes the following steps: Step S21: Load the first collision recognition model in the local runtime environment through the vehicle-mounted remote information control terminal on the vehicle end, monitor the real-time collected vehicle operation data, obtain real-time feature vectors, and determine whether there is a potential collision event based on the real-time feature vectors.

[0064] It should be noted that by loading the first collision recognition model in the local runtime environment through the vehicle-mounted remote information control terminal, the vehicle operation data collected in real time is monitored to obtain real-time feature vectors. Based on the real-time feature vectors, it is determined whether there is a potential collision event. It can use the local computing power of the vehicle to achieve low-latency real-time collision screening and preliminary judgment. It can complete the verification of event triggering conditions without relying on network interaction, ensuring that potential collision events are captured in time and the subsequent data upload process is initiated. At the same time, local filtering reduces the transmission of invalid data.

[0065] Furthermore, step S21 specifically includes the following steps: The first collision recognition model is loaded in the local runtime environment through the vehicle-mounted remote information control terminal, and the raw sensor data is periodically read from the vehicle controller local area network bus. Real-time feature vectors are extracted from the raw sensor data through a sliding window mechanism. The real-time feature vector is input into the first collision recognition model for inference calculation. When the collision probability value output by the first collision recognition model exceeds a preset probability threshold, a potential collision event is identified.

[0066] It is understood that the vehicle-mounted remote information control terminal loads the first collision recognition model in the local runtime environment and periodically reads raw sensor data from the vehicle controller LAN bus. Real-time feature vectors are extracted from the raw sensor data through a sliding window mechanism and input into the first collision recognition model for inference calculation. When the collision probability value output by the first collision recognition model exceeds a preset probability threshold, a potential collision event is identified. This can achieve low-latency continuous real-time monitoring by utilizing the vehicle's local computing power. By capturing data time-series features through a sliding window and filtering normal fluctuations by combining probability thresholds, potential collision events are captured and triggered in a timely and accurate manner.

[0067] Step S22: When the potential collision event is detected, acquire target vehicle data within a preset time period before and after the time of the potential collision event, and send the target vehicle data to the cloud.

[0068] It is understandable that when a potential collision event is identified, target vehicle data within a preset time period before and after the potential collision event can be obtained and sent to the cloud. This can accurately capture key contextual data before and after the collision based on the event triggering mechanism, ensuring the complete data chain required for accident analysis, while avoiding the waste of traffic caused by continuous uploading of all data, thus achieving a balance between data integrity and communication efficiency.

[0069] Furthermore, step S22 specifically includes the following steps: When the potential collision event is detected, the vehicle terminal calls the circular buffer maintained in memory to extract a continuous data segment centered on the trigger time, extending forward for a first preset time and backward for a second preset time as the target vehicle data. After the target vehicle data is formatted, compressed, and encrypted, a communication connection with the cloud is established through a mobile communication module. The processed target vehicle data is uploaded to the cloud.

[0070] It should be understood that by using the circular buffer maintained in memory on the vehicle side to extract continuous data segments before and after the trigger moment, and then encapsulating, compressing, and encrypting the target vehicle data before uploading it to the cloud, the circular buffer mechanism can ensure that key historical data before the collision is not lost due to system power failure or trigger delay, thus achieving accurate capture of the complete data chain before and after the accident. At the same time, compression and encryption optimize transmission bandwidth usage and protect data privacy and security, ensuring that the cloud can receive complete and reliable analysis data.

[0071] Furthermore, the target vehicle data includes at least one of the following: raw airbag controller signal, wheel speed signal, vehicle stability system status signal, and GPS location information.

[0072] It should be noted that the target vehicle data includes at least one of the following: the original signal of the airbag controller, the wheel speed signal, the vehicle stability system status signal, and the location information of the global positioning system. It can provide key data support from multiple dimensions such as collision impact intensity, vehicle motion state, vehicle stability, and accident geographical location, ensuring that the secondary analysis in the cloud has sufficient multi-source evidence.

[0073] Furthermore, the step of uploading the processed target vehicle data to the cloud specifically includes the following steps: The current network status is monitored during the process of uploading the processed target vehicle data to the cloud. When an upload interruption is detected, the current breakpoint position is recorded, and the upload continues from the current breakpoint position after the network is restored; When an upload failure is detected, a preset number of retries will be performed until the upload is successful or the maximum number of retries is reached.

[0074] It is understandable that the system monitors the current network status during the process of uploading the processed target vehicle data to the cloud, records the breakpoint location when an upload interruption is detected and resumes transmission after the network is restored, and performs a preset number of retry operations when an upload failure is detected. This can adapt to the unstable network signal in vehicle movement scenarios, ensure that critical accident data is transmitted to the cloud completely and reliably, and avoid data loss due to network fluctuations.

[0075] This embodiment, through the above-described scheme, loads the first collision recognition model in the local runtime environment via the vehicle-mounted remote information control terminal, monitors the real-time collected vehicle operation data, obtains real-time feature vectors, and determines whether a potential collision event exists based on the real-time feature vectors. When a potential collision event is identified, target vehicle data within a preset time period before and after the time of the potential collision event is acquired and sent to the cloud. This enables low-latency real-time collision screening and preliminary judgment through local loading of the model on the vehicle, ensuring that potential collision events are captured in a timely manner and are not affected by network conditions. At the same time, based on the event triggering mechanism, it accurately captures key context data before and after the collision and uploads it to the cloud, avoiding the waste of traffic caused by continuous uploading of full data, achieving a balance between data integrity and communication efficiency, and providing a reliable basis for in-depth cloud analysis.

[0076] Furthermore, Figure 5 This is a flowchart illustrating the fourth embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention, as shown below. Figure 5 As shown, based on the first embodiment, a fourth embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention is proposed. In this embodiment, step S30 specifically includes the following steps: Step S31: Receive the target vehicle data through the secure communication interface of the cloud, and decrypt, decompress and verify the integrity of the target vehicle data.

[0077] It should be noted that receiving the target vehicle data through the secure communication interface in the cloud, and decrypting, decompressing, and verifying the integrity of the target vehicle data, can ensure the confidentiality and security of the data during transmission and prevent privacy leaks or malicious tampering.

[0078] Step S32: Use the second collision recognition model to perform a second in-depth analysis on the verified target vehicle data to obtain the collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model and is used to identify the collision severity level, collision type, damaged parts, and occupant injury risk.

[0079] Understandably, the second collision recognition model is used to perform a second in-depth analysis on the verified target vehicle data to obtain collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model. It is used to identify the severity level of the collision, the type of collision, the damaged parts, and the risk of injury to the occupants. It can use the high-performance computing power of the cloud to run a high-complexity model to perform in-depth verification and multi-dimensional evaluation of suspected events reported by the vehicle, effectively reducing the false alarm rate and obtaining detailed accident level and damage information.

[0080] Step S33: Generate corresponding early warning information based on the collision analysis results and provide feedback.

[0081] It should be understood that generating corresponding early warning information based on the collision analysis results and providing feedback can transform the in-depth analysis data in the cloud into visualized business instructions or alarms, and push them to user terminals, after-sales platforms or rescue organizations in a timely manner through the feedback mechanism, so as to ensure that the corresponding rescue or service process can be quickly initiated after an accident occurs.

[0082] Furthermore, step S33 specifically includes the following steps: Based on the collision analysis results, a visualized early warning message is generated, which includes vehicle location information, accident time, collision level, and suggested rescue measures. The warning information is pushed to the after-sales service platform, emergency rescue call center or user terminal in real time through the application programming interface.

[0083] Understandably, based on the collision analysis results, a visualized early warning message containing vehicle location information, accident time, collision level, and suggested rescue measures is generated. This message is then pushed in real time to the after-sales service platform, emergency rescue call center, or user terminal via an application programming interface. This transforms abstract collision analysis data into intuitive instructions that can be executed by multiple parties, ensuring that rescue organizations, after-sales departments, and users are simultaneously informed of accident details and suggested measures. This enables real-time cross-platform linkage and precise rescue, minimizing accident response time and improving rescue efficiency.

[0084] This embodiment, through the above-described scheme, receives target vehicle data via a secure communication interface in the cloud, decrypts, decompresses, and verifies the integrity of the target vehicle data; it then uses a second collision recognition model to perform a secondary in-depth analysis on the verified target vehicle data to obtain collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model and is used to identify the collision severity level, collision type, damaged parts, and occupant injury risk. Based on the collision analysis results, corresponding warning information is generated and fed back. This approach ensures the security and integrity of data transmission by decrypting and verifying the received data, and compensates for the vehicle-side computing power limitations by using a highly complex cloud model for secondary in-depth analysis. This significantly reduces the false alarm rate and obtains detailed information such as collision level, damaged parts, and occupant risk, ultimately generating accurate warning information and feeding it back to relevant business platforms. This achieves a secure closed loop from data reception to business linkage, improving accident response efficiency and the targeting of rescue efforts.

[0085] Furthermore, Figure 6 This is a flowchart illustrating the fifth embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention, as shown below. Figure 6 As shown, based on the first embodiment, a fifth embodiment of the cloud-based collaborative vehicle collision recognition and warning method of the present invention is proposed. In this embodiment, after step S30, the cloud-based collaborative vehicle collision recognition and warning method further includes the following steps: Step S40: Obtain the accident data that has been manually or automatically confirmed in the collision analysis results, and add the accident data as a new labeled sample to the historical training dataset.

[0086] It should be noted that by acquiring accident data that has been manually or automatically confirmed in the collision analysis results and adding the accident data as new labeled samples to the historical training dataset, the training data sample library can be continuously enriched using real accident cases.

[0087] Step S50: Retrain and optimize the first collision recognition model based on the updated historical training dataset to generate an updated first collision recognition model.

[0088] Understandably, retraining and optimizing the first collision recognition model based on the updated historical training dataset to generate an updated first collision recognition model can iteratively adjust the model parameters using newly included real accident samples, eliminate the old model's adaptation bias to new scenarios, and enable the vehicle-side model to continuously adapt to new collision features and road conditions, maintaining the timeliness and accuracy of recognition performance.

[0089] Step S60: Send the updated first collision recognition model to the vehicle via the communication network.

[0090] It should be understood that the updated first collision recognition model can be sent to the vehicle via a communication network, enabling remote wireless iterative deployment of the model version and ensuring that the vehicle always loads a model with the latest recognition capabilities.

[0091] In the specific implementation, see Figure 7 , Figure 7 This is a block diagram of the overall system architecture corresponding to the cloud-based collaborative vehicle collision recognition and early warning method of the present invention, as shown below. Figure 7 As shown, this system mainly includes the following four core components: Cloud-based training and model deployment: Based on massive amounts of historical vehicle and accident data, a lightweight model for preliminary judgment is trained in the cloud and deployed to the vehicle's T-BOX unit via OTA (Over-The-Air) technology.

[0092] Real-time vehicle monitoring: The lightweight model deployed on T-BOX analyzes the vehicle sensor data stream in real time to perform preliminary identification and marking of potential collision events in a low-power and high-efficiency manner.

[0093] Collision data acquisition and upload: Once the vehicle model determines that a collision may occur, T-BOX will immediately acquire multi-dimensional, high-precision vehicle signals within a specific time window before and after the collision event (such as the first 15 seconds and the last 5 seconds), and package and encrypt them for upload to the cloud server.

[0094] Precise cloud-based analysis and early warning: After receiving data packets uploaded from the vehicle, the complex large-scale model deployed in the cloud performs in-depth analysis to accurately determine the severity level of the collision, the location of the collision, the possibility of rollover, and other detailed information, and finally generates an early warning, which is pushed to the after-sales service platform or emergency rescue center.

[0095] Specifically: 1) Cloud-based training and model delivery Step 1: Data Acquisition and Preprocessing enter: The real-time data streams from the entire fleet of vehicles reported via the Controller Area Network (CAN) bus include, but are not limited to: vehicle speed, longitudinal / lateral acceleration, yaw rate, steering wheel angle, brake pedal status, and raw signals from the airbag control unit (ACU).

[0096] The historical accident database contains confirmed accident cases, crash test data, insurance claim data, and other information, along with tags such as accident level and collision type.

[0097] Processing rules: Data cleaning: Remove outliers, imput missing values ​​(e.g., using forward and backward mean or linear interpolation) and align timestamps in the input data stream to ensure data quality.

[0098] Feature engineering: Extracting or constructing highly relevant features from the raw signal, such as the rate of change of acceleration (Jerk), velocity vector change, steering wheel angular velocity, etc., to enhance the model's sensitivity to collision events.

[0099] Data annotation: By combining historical accident databases and collision test data, the preprocessed data segments are annotated. The annotation information includes event levels such as "non-collision", "minor collision", "moderate collision", and "severe collision".

[0100] Output: High-quality, labeled feature datasets that can be used for model training.

[0101] Step 2: Lightweight Model Training for Vehicle End enter: The feature dataset output from step 1.

[0102] Processing rules: Model selection: Select machine learning algorithms with low computational complexity and small memory footprint, such as decision trees, support vector machines (SVM), or small neural networks (e.g., containing 1-2 hidden layers, each with 8-16 neurons) to adapt to the hardware resource constraints of the in-vehicle telematics box (T-BOX).

[0103] Model training: The selected model is trained using the feature dataset with the goal of enabling it to distinguish between normal driving behavior and potential collision events.

[0104] The optimization goal is high recall, which means it is better to have false positives than false negatives, ensuring that all suspicious events are captured.

[0105] Model quantization and compression: The trained model is quantized (e.g., 8-bit integer (quantization format) (8-bit Integer, INT8) quantization) and pruned to further reduce the model size and computational overhead, ensuring efficient operation on T-BOX.

[0106] Output: A compressed and optimized lightweight preliminary collision assessment model file suitable for vehicle deployment.

[0107] Step 3: Over-The-Air (OTA) Model Distribution and Deployment enter: The lightweight model file output from step 2.

[0108] List of target vehicles.

[0109] Processing rules: Packaging and Encryption: The model files are packaged into OTA upgrade packages and then encrypted and signed to ensure the security of transmission and deployment.

[0110] OTA Push: Securely push the upgrade package to the target vehicle's T-BOX via a cloud-based OTA platform.

[0111] Dynamic loading: After receiving the upgrade package, T-BOX verifies and decrypts it, and dynamically loads the new model into its runtime environment to replace or update the old model. The entire process has no impact on the normal operation of the vehicle.

[0112] Output: A lightweight collision assessment model was successfully deployed and run on the target vehicle's T-BOX.

[0113] 2) Real-time vehicle monitoring Step 1: Real-time signal flow monitoring enter: Raw data streams generated in real time by vehicle sensors (such as accelerometers, gyroscopes, etc.).

[0114] Processing rules: Sliding window processing: Set a short-duration sliding window (e.g., 100 milliseconds) to continuously collect sensor data within the window.

[0115] Real-time feature extraction: Within each window, feature values ​​(such as peak acceleration, rate of change, etc.) that match the input of the lightweight model are calculated in real time.

[0116] Output: A continuously generated, real-time feature vector stream that can be used by the model for judgment.

[0117] Step 2: Preliminary assessment of the collision event enter: The real-time feature vector stream output from step 1.

[0118] A lightweight model deployed on T-BOX.

[0119] Judgment rules: Model inference: Input real-time feature vectors into a lightweight model for inference calculation.

[0120] Threshold determination: The model outputs a collision probability value or classification result. Set a predefined confidence threshold (e.g., probability > 0.7).

[0121] When the model output exceeds this threshold, it is considered a potential collision event.

[0122] Output: A "potential collision event" trigger signal includes a timestamp of the event occurring.

[0123] 3) Collision data collection and uploading Step 1: High-precision data caching and acquisition enter: The "potential collision event" trigger signal output in step 2.

[0124] High-frequency, multi-dimensional data streams on the vehicle data bus (e.g., raw waveform data from the ACU, speeds of each wheel, status of the Anti-lock Braking System (ABS) / Electronic Stability Program (ESP), gear information, GPS location, etc.).

[0125] Processing rules: Circular Buffer: T-BOX maintains a circular buffer in memory to continuously cache high-precision vehicle data for a recent period of time (such as 30 seconds).

[0126] Data Freeze and Acquisition: Upon receiving a trigger signal, the system extracts data from the buffer for the first 15 seconds and the last 5 seconds, centered on the trigger timestamp. This "more data before and less after" strategy aims to comprehensively record the vehicle dynamics before the collision, as well as the instantaneous and short-term consequences of the collision.

[0127] Output: A temporary data packet containing high-precision, multi-dimensional data for 20 seconds before and after the collision event.

[0128] Step 2: Data Packaging and Uploading enter: The temporary data packet output in step 1.

[0129] Processing rules: Data Packaging: The collected data, along with metadata such as Vehicle Identification Number (VIN) code, event timestamp, GPS location, and preliminary judgment results of the lightweight model, are packaged into a structured data file (such as JSON or Protobuf format).

[0130] Compression and encryption: Data packets are compressed to reduce transmission size, and asymmetric encryption and other methods are used to ensure the confidentiality and integrity of data during the upload process.

[0131] Reliable Upload: Upload encrypted data packets to the designated cloud server via 4G / 5G network. The upload mechanism should have the functions of resuming interrupted uploads and retrying on failure to cope with unstable network conditions.

[0132] 4) Precise cloud-based analysis and early warning Step 1: Precise Analysis of Large Cloud Models enter: Collision event data packets uploaded from the vehicle.

[0133] Processing rules: Data parsing and verification: After receiving the data packet, the cloud server decrypts, decompresses, and verifies the format.

[0134] Deep model analysis: The parsed, high-precision data is input into a complex analytical model deployed in the cloud. This model can be a deep neural network (such as a Long Short-Term Memory (LSTM) network or a Transformer), capable of analyzing complex correlations in time-series data. The model will make accurate judgments on the following: Collision severity level: Accidents are precisely classified into levels based on energy changes, deformation estimates, etc. (e.g., Event Data Recorder (EDR) levels 1-4).

[0135] Collision type and location: Analyze multi-axis acceleration and yaw rate data to determine whether it is a frontal, side, rear-end, or rollover collision.

[0136] Occupant injury risk assessment: Assess the risk of injury to occupants by combining information such as whether seat belts are worn and whether airbags have deployed.

[0137] Output: A detailed collision analysis report includes structured information such as collision level, type, location, and occupant risk.

[0138] Step 2: Early Warning Generation and Push Notification enter: The collision analysis report output in step 1.

[0139] Processing rules: Decision engine: Based on the collision severity level in the analysis report, different business logic is triggered. For example, a severe collision automatically triggers the highest priority emergency rescue process.

[0140] Warning information generation: Generates visual warning information containing key information such as vehicle information (VIN, model), accident time, location (GPS), and collision analysis summary.

[0141] Multi-channel push: Through the Application Programming Interface (API), the warning information is pushed in real time to downstream applications such as the OEM's after-sales service platform, Customer Relationship Management (CRM) system, and emergency call center (e-Call).

[0142] Output: Real-time collision warnings have been sent to relevant business terminals.

[0143] This triggers the corresponding after-sales care or emergency rescue process.

[0144] This embodiment, through the above-described scheme, acquires accident data from the collision analysis results that have been manually or automatically confirmed, and adds this accident data as new labeled samples to the historical training dataset; based on the updated historical training dataset, it retrains and optimizes the first collision recognition model to generate an updated first collision recognition model; and sends the updated first collision recognition model to the vehicle end through a communication network. This enables the continuous enrichment of the training sample library through a closed-loop feedback mechanism of real accident data, and utilizes the newly included collision features to iteratively optimize and remotely deploy the vehicle-end model, eliminating the adaptive bias caused by outdated data, ensuring that the vehicle end always has a high-precision recognition capability that adapts to the latest operating conditions, and achieving continuous evolution and long-term maintenance of model performance.

[0145] Accordingly, the present invention further provides a cloud-based collaborative vehicle collision recognition and early warning device.

[0146] Reference Figure 8 , Figure 8 This is a functional block diagram of the first embodiment of the cloud-based collaborative vehicle collision recognition and early warning device of the present invention.

[0147] In the first embodiment of the cloud-based collaborative vehicle collision recognition and warning device of the present invention, the cloud-based collaborative vehicle collision recognition and warning device includes: The model adaptation module 10 is used to train a first collision recognition model adapted to the vehicle's hardware resources via the cloud, and to send the first collision recognition model to the vehicle via a communication network.

[0148] The data acquisition module 20 is used to monitor the real-time vehicle operation data by running the first collision recognition model on the vehicle end. When a potential collision event is identified, it acquires target vehicle data within a preset time period before and after the time of the potential collision event and sends the target vehicle data to the cloud.

[0149] The analysis and early warning module 30 is used to receive the target vehicle data through the cloud, perform secondary analysis on the target vehicle data using the second collision recognition model to obtain collision analysis results, and generate early warning information based on the collision analysis results.

[0150] The steps for implementing each functional module of the cloud-based collaborative vehicle collision recognition and warning device can be referred to in the various embodiments of the cloud-based collaborative vehicle collision recognition and warning method of the present invention, and will not be repeated here.

[0151] Furthermore, this embodiment of the invention also proposes a storage medium storing a cloud-based collaborative vehicle collision recognition and warning program. When the cloud-based collaborative vehicle collision recognition and warning program is executed by a processor, it implements the operations described in the above-described embodiment of the cloud-based collaborative vehicle collision recognition and warning method.

[0152] Those skilled in the art will understand that all or part of the steps in the methods described above can be implemented by a program instructing related hardware. The program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium is a computer-readable storage medium, including: USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program code.

[0153] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0154] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0155] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A cloud-based collaborative vehicle collision recognition and early warning method, characterized in that, The cloud-based collaborative vehicle collision recognition and early warning method includes: A first collision recognition model adapted to vehicle-side hardware resources is trained in the cloud and then sent to the vehicle-side via a communication network. The vehicle terminal runs the first collision recognition model to monitor the real-time collected vehicle operation data. When a potential collision event is identified, the target vehicle data within a preset time period before and after the time of the potential collision event is obtained and the target vehicle data is sent to the cloud. The system receives the target vehicle data via the cloud, performs secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generates warning information based on the collision analysis results.

2. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 1, characterized in that, The process of training a first collision recognition model adapted to vehicle-side hardware resources via the cloud and sending the first collision recognition model to the vehicle-side via a communication network includes: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, a first collision recognition model adapted to vehicle-side hardware resources is trained based on the historical operational data and the accident case data. The first collision recognition model is sent to the vehicle via over-the-air download through a communication network.

3. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 2, characterized in that, The process of aggregating historical operational data and confirmed accident case data of the entire fleet via the cloud, and training a first collision recognition model adapted to vehicle-side hardware resources based on the historical operational data and the accident case data, includes: By aggregating historical operational data and confirmed accident case data of the entire fleet in the cloud, an initial collision recognition model is trained using deep learning algorithms based on the historical operational data and the accident case data. The computing power capacity and storage space limitations of the vehicle-mounted remote control terminal are obtained. The initial collision recognition model is compressed and optimized through model pruning and parameter quantization to generate the first collision recognition model.

4. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 2, characterized in that, Sending the first collision recognition model to the vehicle via over-the-air download through a communication network includes: The first collision recognition model is packaged into a remote upgrade package with encrypted signature processing via the cloud; The remote upgrade package is pushed to the vehicle via an over-the-air download through a communication network.

5. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 1, characterized in that, The process of monitoring real-time vehicle operation data by running the first collision recognition model on the vehicle end, and when a potential collision event is identified, acquiring target vehicle data within a preset time period before and after the time of the potential collision event, and sending the target vehicle data to the cloud includes: The vehicle-mounted remote information control terminal loads the first collision recognition model in the local runtime environment, monitors the real-time collected vehicle operation data, obtains real-time feature vectors, and determines whether there is a potential collision event based on the real-time feature vectors. When a potential collision event is detected, target vehicle data within a preset time period before and after the time of the potential collision event is obtained, and the target vehicle data is sent to the cloud.

6. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 5, characterized in that, The step of loading the first collision recognition model in the local runtime environment through the vehicle-mounted remote information control terminal, monitoring the real-time collected vehicle operation data, obtaining real-time feature vectors, and determining whether a potential collision event exists based on the real-time feature vectors includes: The first collision recognition model is loaded in the local runtime environment through the vehicle-mounted remote information control terminal, and the raw sensor data is periodically read from the vehicle controller local area network bus. Real-time feature vectors are extracted from the raw sensor data through a sliding window mechanism. The real-time feature vector is input into the first collision recognition model for inference calculation. When the collision probability value output by the first collision recognition model exceeds a preset probability threshold, a potential collision event is identified.

7. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 5, characterized in that, The step of acquiring target vehicle data within a preset time period before and after the potential collision event is detected, and sending the target vehicle data to the cloud, includes: When the potential collision event is detected, the vehicle terminal calls the circular buffer maintained in memory to extract a continuous data segment centered on the trigger time, extending forward for a first preset time and backward for a second preset time as the target vehicle data. After the target vehicle data is formatted, compressed, and encrypted, a communication connection with the cloud is established through a mobile communication module. The processed target vehicle data is uploaded to the cloud.

8. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 7, characterized in that, The target vehicle data includes at least one of the following: raw airbag controller signal, wheel speed signal, vehicle stability system status signal, and GPS location information.

9. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 7, characterized in that, Uploading the processed target vehicle data to the cloud includes: The current network status is monitored during the process of uploading the processed target vehicle data to the cloud. When an upload interruption is detected, the current breakpoint position is recorded, and the upload continues from the current breakpoint position after the network is restored; When an upload failure is detected, a preset number of retries will be performed until the upload is successful or the maximum number of retries is reached.

10. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 1, characterized in that, The process of receiving target vehicle data via the cloud, performing secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generating warning information based on the collision analysis results includes: The target vehicle data is received through the secure communication interface of the cloud, and the target vehicle data is decrypted, decompressed, and its integrity is verified. The second collision recognition model is used to perform a second in-depth analysis on the verified target vehicle data to obtain the collision analysis results. The second collision recognition model has a higher computational complexity than the first collision recognition model and is used to identify the collision severity level, collision type, damaged parts and occupant injury risk. Based on the collision analysis results, corresponding early warning information is generated and fed back.

11. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 10, characterized in that, The process of generating and feeding back corresponding early warning information based on the collision analysis results includes: Based on the collision analysis results, a visualized early warning message is generated, which includes vehicle location information, accident time, collision level, and suggested rescue measures. The warning information is pushed to the after-sales service platform, emergency rescue call center or user terminal in real time through the application programming interface.

12. The cloud-based collaborative vehicle collision recognition and early warning method as described in claim 1, characterized in that, After receiving the target vehicle data via the cloud, performing secondary analysis on the target vehicle data using a second collision recognition model to obtain collision analysis results, and generating warning information based on the collision analysis results, the cloud-based collaborative vehicle collision recognition and warning method further includes: Obtain accident data from the collision analysis results that have been manually or automatically confirmed, and add the accident data as new labeled samples to the historical training dataset; The first collision recognition model is retrained and optimized based on the updated historical training dataset to generate an updated first collision recognition model. The updated first collision recognition model is sent to the vehicle via a communication network.

13. A cloud-based collaborative vehicle collision recognition and early warning device, characterized in that, The cloud-based collaborative vehicle collision recognition and early warning device includes: The model adaptation module is used to train a first collision recognition model adapted to the vehicle's hardware resources via the cloud, and send the first collision recognition model to the vehicle via a communication network. The data acquisition module is used to monitor the real-time vehicle operation data by running the first collision recognition model on the vehicle end. When a potential collision event is identified, the module acquires target vehicle data within a preset time period before and after the time of the potential collision event and sends the target vehicle data to the cloud. The analysis and early warning module is used to receive the target vehicle data through the cloud, perform secondary analysis on the target vehicle data using the second collision recognition model to obtain collision analysis results, and generate early warning information based on the collision analysis results.

14. A cloud-based collaborative vehicle collision recognition and early warning device, characterized in that, The cloud-based collaborative vehicle collision recognition and warning device includes: a memory, a processor, and a cloud-based collaborative vehicle collision recognition and warning program stored in the memory and executable on the processor, wherein the cloud-based collaborative vehicle collision recognition and warning program is configured to implement the steps of the cloud-based collaborative vehicle collision recognition and warning method as described in any one of claims 1 to 12.

15. A storage medium, characterized in that, The storage medium stores a cloud-based collaborative vehicle collision recognition and warning program, which, when executed by a processor, implements the steps of the cloud-based collaborative vehicle collision recognition and warning method as described in any one of claims 1 to 12.

Images