Vehicle collision detection method, device and storage medium

Abstract

The application provides a vehicle collision detection method, device and storage medium. It relates to the technical field of data processing. The method comprises: acquiring multi-source sensing information of a vehicle during driving, the multi-source sensing information comprising first environment information, first state information of passengers and first collision information; determining a target collision parameter according to the first environment information, the first state information and the first collision information, the target collision parameter being used to determine whether a road collision event occurs to the vehicle; determining a vehicle control parameter according to the target collision parameter, the vehicle control parameter being used to control the vehicle to perform a corresponding driving operation; and when the target collision parameter indicates that the road collision event occurs, sending road collision information of the vehicle to a cloud server, so that the vehicle drives based on a control instruction issued by the cloud server. The reliability of collision detection of the vehicle is improved.

Description

Technical Field

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

[0002] In the field of intelligent driving assistance, with the popularization of autonomous driving functions, autonomous vehicles need to complete driving tasks independently without driver intervention during operation, ensuring smooth driving and operational safety.

[0003] In related technologies, vehicles typically use environmental perception for collision detection, which involves collecting information about surrounding obstacles based on cameras, lidar, etc., and then using identification and trajectory prediction to provide a collision warning and execute corresponding driving operations after a collision occurs.

[0004] However, in the above process, in scenarios where perception is limited, such as severe weather or sudden long tails, relying solely on environmental perception information for collision judgment can easily lead to misjudgments or omissions, such as minor scratches at low speeds or collisions with low visibility, resulting in low reliability of vehicle collision detection. Summary of the Invention

[0005] This application provides a vehicle collision detection method, device, and storage medium to solve the problem of low reliability in vehicle collision detection.

[0006] Firstly, this application provides a vehicle collision detection method, including:

[0007] Acquire multi-source perception information of the vehicle during its driving process. The multi-source perception information includes first environmental information, first passenger state information, and first collision information.

[0008] Based on the first environmental information, the first state information, and the first collision information, the target collision parameters are determined. The target collision parameters are used to determine whether a road collision event has occurred.

[0009] Based on the target collision parameters, the vehicle control parameters are determined, and the vehicle control parameters are used to control the vehicle to perform corresponding driving operations.

[0010] When the target collision parameters indicate that a road collision event has occurred, the vehicle's road collision information is sent to the cloud server so that the vehicle can drive based on the control commands issued by the cloud server.

[0011] In one possible implementation, obtaining the first collision information includes:

[0012] Acquire collision signals collected by detection devices in various areas of the vehicle body;

[0013] The collision signal is preprocessed and the region is located to obtain the corresponding collision result and collision location information, and the collision result and collision location information are determined as the first collision information.

[0014] In one possible implementation, determining the target collision parameters includes:

[0015] Alignment processing is performed on the first environmental information, the first state information, and the first collision information to obtain the second environmental information, the second state information, and the second collision information;

[0016] Based on the second environmental information, the second state information, and the second collision information, the target collision parameters are determined.

[0017] In one possible implementation, obtaining the second environmental information, the second state information, and the second collision information includes:

[0018] The first environmental information, the first state information, and the first collision information are time-stamped to obtain multi-source sensing information at the same moment.

[0019] Multi-source perception information at the same moment is mapped to the same vehicle coordinate system to generate second environmental information, second state information and second collision information.

[0020] In one possible implementation, determining the target collision parameters based on the second environmental information, the second state information, and the second collision information includes:

[0021] The second environmental information, the second state information, and the second collision information are input into the prediction model to predict the target collision parameters.

[0022] In one possible implementation, determining the target collision parameters includes:

[0023] Determine the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information;

[0024] The target collision parameters are determined based on the first collision parameters, the second collision parameters, and the third collision parameters.

[0025] In one possible implementation, determining the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information includes:

[0026] The first environmental information is input into the first model to obtain the first collision parameters;

[0027] The first state information is input into the second model to obtain the second collision parameters;

[0028] The first collision information is input into the third model to obtain the third collision parameters.

[0029] In one possible implementation, determining the target collision parameters based on the first collision parameters, the second collision parameters, and the third collision parameters includes:

[0030] Determine the collision weights corresponding to each collision parameter;

[0031] The target collision parameters are calculated based on the first collision parameter, the second collision parameter, the third collision parameter and the corresponding collision weights.

[0032] In one possible implementation, determining vehicle control parameters based on target collision parameters includes:

[0033] When the target collision parameter is greater than the collision threshold, the vehicle control parameter is determined as the first control parameter used to control the vehicle to perform deceleration and stopping operations.

[0034] In one possible implementation, determining vehicle control parameters based on target collision parameters includes:

[0035] When the target collision parameter is less than or equal to the collision threshold, the vehicle control parameter is determined as a second control parameter used to control the vehicle to maintain its current state of motion.

[0036] In one possible implementation, sending road collision information of the vehicle to a cloud server includes:

[0037] Acquire video and vehicle location information at the time of a road collision;

[0038] The video information, vehicle location information, and target collision parameters are sent to the cloud server as road collision information.

[0039] Secondly, this application provides a vehicle collision detection method, including:

[0040] Receive road collision information sent from the vehicle;

[0041] Based on road collision information, a collision control strategy for the vehicle is determined, which is used to control the vehicle to perform corresponding driving operations.

[0042] Based on the collision control strategy, control commands are generated and sent to the vehicle so that the vehicle can drive according to the control commands.

[0043] In one possible implementation, a collision control strategy for the vehicle is determined based on road collision information, including:

[0044] When it is determined that no road collision event has occurred based on road collision information, the collision control strategy is determined to be a recovery strategy used to control the vehicle to resume normal driving.

[0045] In one possible implementation, a collision control strategy for the vehicle is determined based on road collision information, including:

[0046] When a road collision event is determined based on road collision information, the collision control strategy is determined to be a handling strategy used to push road collision information to the platform.

[0047] In one possible implementation, after receiving road collision information sent by the vehicle, the method further includes:

[0048] The road collision information is verified to obtain the verification results;

[0049] If the verification result is that the verification fails, the verification result is sent to the vehicle terminal so that the vehicle terminal can perform a data retransmission operation.

[0050] Thirdly, this application provides a vehicle collision detection device, wherein:

[0051] The acquisition module is used to acquire multi-source perception information of the vehicle during its driving process. The multi-source perception information includes first environmental information, first passenger state information, and first collision information.

[0052] The first determining module is used to determine the target collision parameters based on the first environmental information, the first state information and the first collision information. The target collision parameters are used to determine whether the vehicle has caused a road collision event.

[0053] The second determining module is used to determine vehicle control parameters based on the target collision parameters. The vehicle control parameters are used to control the vehicle to perform corresponding driving operations.

[0054] The sending module is used to send the vehicle's road collision information to the cloud server when the target collision parameters indicate that a road collision event has occurred, so that the vehicle can drive based on the control commands issued by the cloud server.

[0055] Fourthly, this application provides a vehicle collision detection device, wherein:

[0056] The receiving module is used to receive road collision information sent by the vehicle.

[0057] The third determining module is used to determine the vehicle's collision control strategy based on the road collision information. The collision control strategy is used to control the vehicle to perform corresponding driving operations.

[0058] The generation module is used to generate control commands based on the collision control strategy and issue control commands to the vehicle so that the vehicle can drive based on the control commands.

[0059] Fifthly, embodiments of this application provide a vehicle collision detection device, comprising: at least one processor and a memory; the memory storing computer execution instructions; the at least one processor executing the computer execution instructions stored in the memory, causing the at least one processor to perform the vehicle collision detection method as described in the first aspect and any one of the first aspects.

[0060] In a sixth aspect, embodiments of this application provide a vehicle collision detection device, comprising: at least one processor and a memory; the memory storing computer execution instructions; and at least one processor executing the computer execution instructions stored in the memory, causing the at least one processor to perform the vehicle collision detection method as described in the second aspect above and any one of the second aspects.

[0061] In a seventh aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the vehicle collision detection method as described in the first aspect and any one of the first aspects.

[0062] Eighthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the vehicle collision detection method as described in the second aspect above and any one of the second aspects.

[0063] The vehicle collision detection method, device, and storage medium provided in this application acquire environmental information, passenger status information, and collision detection information during vehicle operation. They then fuse this multi-source perception information to determine target collision parameters, and based on these parameters, determine vehicle control parameters. When the target collision parameters indicate a road collision event, they send road collision information to the cloud and receive control commands from the cloud. This method, by introducing multi-source fusion judgment of physical collision detection information, environmental perception information, and passenger status information, can accurately identify low-speed minor scratches or low-visibility collisions in scenarios with limited perception, such as severe weather or sensor obstruction. This effectively avoids the problem of missed or false judgments caused by the failure of single environmental perception in complex scenarios. It can autonomously detect and handle collision events without driver intervention. Furthermore, through a vehicle-cloud collaborative confirmation mechanism, the vehicle can drive based on control commands issued from the cloud, achieving closed-loop management of collision event handling. This improves the reliability of vehicle collision detection and ensures the operational safety of autonomous vehicles. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0065] Figure 1 This is a schematic diagram illustrating an application scenario provided in the embodiments of this application;

[0066] Figure 2 A schematic flowchart of a vehicle collision detection method provided in an embodiment of this application;

[0067] Figure 3 This is a schematic diagram illustrating the deployment location of the vehicle-side sensor detection provided in an embodiment of this application.

[0068] Figure 4 A schematic diagram illustrating a possible process for determining target collision parameters, provided in an embodiment of this application;

[0069] Figure 5 A schematic diagram illustrating a possible process for determining target collision parameters, provided in an embodiment of this application;

[0070] Figure 6 A schematic diagram illustrating the process of determining vehicle control parameters provided in an embodiment of this application;

[0071] Figure 7 A flowchart illustrating another vehicle collision detection method provided in this application embodiment;

[0072] Figure 8 A schematic diagram of the system architecture provided for embodiments of this application;

[0073] Figure 9 A schematic flowchart illustrating another vehicle collision detection method provided in this application embodiment;

[0074] Figure 10 This is a schematic diagram of the structure of a vehicle collision detection device provided in an embodiment of this application;

[0075] Figure 11 This is a schematic diagram of another vehicle collision detection device provided in an embodiment of this application;

[0076] Figure 12 This is a schematic diagram of the structure of the vehicle collision detection device provided in the embodiments of this application.

[0077] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0078] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0079] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0080] Autonomous driving collision detection technology is widely used in safety scenarios for Level 4 and above driverless vehicles. Vehicles need to autonomously complete collision detection and emergency post-processing operations in complex road conditions, scenarios with limited perception, and long-tail traffic situations without driver intervention, to prevent secondary escalation of collisions and scratches, and to ensure the safety of occupants and other road users.

[0081] In related technologies, collision detection for autonomous vehicles typically employs an environmental perception fusion approach. This involves using cameras, millimeter-wave radar, and lidar to collect information about surrounding obstacles, identifying them, and predicting their trajectories to provide collision warnings. After a collision occurs, the vehicle automatically executes pre-set basic safety response procedures. While this solution can achieve basic collision detection functionality in normal scenarios, environmental perception capabilities are prone to degradation in adverse weather conditions, physical obstructions, or algorithm failures. This can lead to ineffective identification of low-speed minor scratches or low-visibility collisions, resulting in missed or false alarms.

[0082] Furthermore, the existing solution does not introduce an independent physical collision detection backup, lacks a remote safety officer collaborative confirmation mechanism after a collision occurs, and can only perform local basic responses, making it impossible to quickly classify and remotely intervene in collision events.

[0083] In summary, the core problem of insufficient reliability in existing technologies for collision detection is that single environmental perception is prone to failure in scenarios with limited perception, leading to missed or false detections of minor scratches at low speeds or collisions with low visibility. Furthermore, the lack of independent physical collision detection backup and vehicle-cloud collaborative confirmation mechanisms makes it difficult to meet the reliability requirements of collision detection in fully autonomous driving scenarios.

[0084] In this embodiment, target collision parameters are obtained by acquiring environmental information, passenger status information and collision detection information. The target collision parameters are compared with the collision threshold to determine vehicle control parameters. When a collision is determined to have occurred, the system reports to the cloud to receive control commands. In the aforementioned method, by introducing a physical collision detection link and fusing it with environmental perception information and passenger status information using confidence weighting, the problem of missed or false judgments caused by the failure of single environmental perception in scenarios with limited perception, such as severe weather or sensor obstruction, is effectively avoided. Furthermore, it can effectively identify low-speed minor scratches or low-visibility collisions. Moreover, through preprocessing with timestamp alignment and coordinate system mapping, the consistency of multi-source information in time and space is ensured, improving the accuracy of fusion judgment. Additionally, by comparing target collision parameters with collision thresholds, differentiated control is achieved by maintaining the vehicle's current motion state when no collision occurs and controlling the vehicle to decelerate and stop during a collision. Furthermore, through two-way communication between the vehicle and the cloud, the cloud can generate recovery or rescue strategies based on road collision information and issue control commands, realizing remote classification and closed-loop handling of collision events, avoiding operational interruptions due to false alarms or the inability to provide timely rescue in the event of a real collision. By improving the reliability of collision detection, the operational safety of fully autonomous vehicles is ensured.

[0085] To facilitate understanding, the following will be combined with... Figure 1 The application scenarios applicable to the embodiments of this application will be described.

[0086] Figure 1 This is a schematic diagram illustrating an application scenario provided in an embodiment of this application. Please refer to [link / reference]. Figure 1 This includes vehicles, vehicle collision detection equipment, and cloud servers. The vehicle collision detection equipment can be deployed in vehicles.

[0087] During the operation of an autonomous vehicle, when the vehicle encounters a scenario with limited perception, multi-source perception devices deployed on the vehicle collect collision detection information, environmental perception information, and passenger status information around the vehicle. Based on the fusion of the above multi-source perception information, the system determines whether a collision event has occurred and the corresponding driving control method. This enables more reliable and safer autonomous driving collision detection and emergency response, improving the robustness and safety of fully autonomous driving scenarios.

[0088] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0089] Figure 2This is a schematic flowchart illustrating a vehicle collision detection method provided in an embodiment of this application. Please refer to [link / reference]. Figure 2 As shown, the method may include the following steps:

[0090] S201. Acquire multi-source perception information of the vehicle during its driving process.

[0091] The execution subject of this application embodiment can be a vehicle or a vehicle collision detection device installed in a vehicle. The vehicle collision detection device can be implemented by software or by a combination of software and hardware.

[0092] The multi-source perception information includes first environmental information, first passenger state information, and first collision information.

[0093] The first environmental information can refer to information about obstacles around the vehicle (e.g., obstacle type, location, speed and direction of movement) and road environment information (e.g., lane lines, curbs and traffic signs) collected by sensors deployed on the vehicle body.

[0094] The first state information can refer to passenger state data, including but not limited to the passenger's seating position, changes in body posture, and passenger displacement information caused by abnormal vehicle movement (e.g., sudden acceleration, sudden deceleration, or impact).

[0095] The first collision information can refer to the physical collision signal collected and preprocessed by collision detection sensors located in key areas of the vehicle body, including the judgment result of whether the vehicle body has made physical contact and the specific location of the vehicle body area where the contact occurred.

[0096] In some embodiments, environmental perception data around the vehicle can be collected in real time by cameras, millimeter-wave radar and lidar deployed on the front, rear and sides of the vehicle body, and target detection and feature extraction can be performed on the collected multi-sensor data to obtain information such as obstacle type, position, speed and direction of movement as the first environmental information.

[0097] In some embodiments, an in-vehicle Occupancy Monitoring System (OMS) can be used to collect image data of each passenger in the cabin in real time, and perform human key point detection and posture analysis on the image data to obtain information on the passenger's seating position and changes in human posture. At the same time, the vehicle's motion state is monitored by the acceleration sensing unit in the OMS system. When the vehicle is detected to have undergone rapid acceleration, rapid deceleration or abnormal impact, the displacement state information of the passengers is recorded as the first state information.

[0098] In some embodiments, the first collision information can be obtained by: acquiring collision signals collected by detection devices in various areas of the vehicle body; preprocessing and locating the collision signals to obtain the corresponding collision results and collision position information, and determining the collision results and collision position information as the first collision information.

[0099] The collision result can refer to the output result after signal conditioning, filtering and threshold comparison of the raw physical signal collected by the collision detection sensor.

[0100] Collision location information can refer to the spatial coordinates or preset area number of the sensor in the vehicle coordinate system, which is determined based on the preset area identifier of the vehicle body corresponding to the sensor that collected the collision signal, and is used to accurately locate the point where the collision occurred.

[0101] In some embodiments, collision detection sensors configured in various areas of the vehicle body collect vibration signals, deformation signals, or pressure signals of the vehicle body structure in real time. The raw analog signals output by the sensors are amplified and filtered by a signal conditioning circuit, then converted into digital signals by an analog-to-digital converter. The collision detection controller performs sliding window filtering and peak detection on the digital signals. When the signal amplitude exceeds a preset collision threshold, it is determined that a collision has occurred in the area where the sensor is located, and the preset area identifier corresponding to the sensor is recorded as the collision location information.

[0102] Below, in conjunction with Figure 3 The deployment locations of the vehicle-side sensors are described in detail.

[0103] Figure 3 This is a schematic diagram illustrating the deployment location of the vehicle-side sensor detection provided in an embodiment of this application. Please refer to... Figure 3 To achieve collision detection without blind spots, the vehicle body can be divided into multiple key collision detection areas. The specific number of sensors needs to be determined based on the vehicle model.

[0104] Assume that the sensors are deployed in 10 key collision detection areas on the vehicle body. Specifically, 4-6 collision detection sensors are placed in the front bumper area for rapid identification of frontal collisions (e.g., collisions with vehicles, pedestrians, obstacles, etc.), with a response time of ≤20ms; 3-6 collision detection sensors are placed in the rear bumper area for identification of rear-end collisions, with a detection range covering the entire width of the rear bumper, capable of identifying minor rear-end collisions at speeds ≤15km / h.

[0105] One or two sensors are arranged in the front left fender area to detect collisions and scrapes in the left front fender area (e.g., side scrapes with roadside guardrails or other vehicles); one or two sensors are arranged in the front right fender area to detect collisions and scrapes in the right front fender area; one or two sensors are arranged in the rear left fender area to detect collisions and scrapes in the left rear fender area; and one or two sensors are arranged in the rear right fender area to detect collisions and scrapes in the right rear fender area.

[0106] One to two sensors are installed in the front left door area to detect side collisions and scrapes (e.g., side collisions, door scrapes) in that area; one to two sensors are installed in the front right door area to detect side collisions and scrapes in that area; one to two sensors are installed in the rear left door area to detect side collisions and scrapes in that area; and one to two sensors are installed in the rear right door area to detect side collisions and scrapes in that area. All sensor signals are centrally processed by the collision detection controller.

[0107] The above method enables collision detection coverage in all directions of the vehicle body, ensuring that any low-speed minor scratches or collisions can be effectively detected, providing reliable physical evidence for subsequent fusion judgment.

[0108] S202. Determine the target collision parameters based on the first environmental information, the first state information, and the first collision information.

[0109] Among them, the target collision parameters are used to determine whether a road collision event has occurred.

[0110] Understandably, the target collision parameter can refer to the comprehensive confidence score output after multi-source information fusion processing of the first environmental information, the first state information, and the first collision information. This score is used to indicate the credibility of a collision event occurring at the current moment, and the value range is usually from 0 to 1. The higher the score, the greater the probability of a collision.

[0111] Optionally, the acquired first environmental information, first state information, and first collision information can be time-stamped and aligned with the spatial coordinate system to unify the three types of information to the same reference time and the same vehicle coordinate system. Subsequently, the confidence scores of the three types of information after alignment are determined to obtain their respective confidence scores. Finally, the three confidence scores are weighted and summed based on the preset fusion weights, and the weighted sum is used as the target collision parameter.

[0112] S203. Determine the vehicle control parameters based on the target collision parameters.

[0113] Among them, vehicle control parameters are used to control the vehicle to perform corresponding driving operations.

[0114] In some embodiments, since the target collision parameter is a continuous comprehensive confidence score, it cannot be directly used to drive the vehicle to perform specific driving actions. Therefore, it needs to be compared with a preset collision threshold. Based on the comparison result, the continuous confidence score is mapped to discrete, executable vehicle control parameters to achieve precise control of the vehicle.

[0115] In some embodiments, the target collision parameters are compared with a preset collision threshold: if the target collision parameters are greater than the collision threshold, it indicates that the overall credibility of the current collision event is high, and the vehicle control parameters may include information such as target deceleration, braking pressure, and target parking position; if the target collision parameters are less than or equal to the collision threshold, it indicates that the overall credibility of the current collision event is low, and the vehicle control parameters may include control commands to maintain the current vehicle speed and driving direction.

[0116] S204. When the target collision parameters indicate that a road collision event has occurred, send the vehicle's road collision information to the cloud server so that the vehicle can drive based on the control commands issued by the cloud server.

[0117] Road collision information can refer to a set of information related to road collision events, including at least the vehicle's external surround view video information, vehicle position information, and target collision parameters at the time of the collision, which is used for remote safety officers in the cloud to conduct secondary manual confirmation and decision-making.

[0118] In some embodiments, road collision information of a vehicle can be sent to a cloud server in the following manner: acquiring video information and vehicle location information at the time of the road collision event; and sending the video information, vehicle location information, and target collision parameters as road collision information to the cloud server.

[0119] Video information can refer to the video stream data of the external environment of the vehicle collected by surround-view cameras deployed in all directions around the vehicle within a preset time period before and after the collision event. It includes real-time images from all directions around the vehicle and is used for cloud-based safety operators to remotely view the actual scene of the collision to help determine the authenticity and severity of the collision event.

[0120] Vehicle location information refers to the spatial coordinate data of the vehicle's current location calculated in real time by the vehicle positioning system, including latitude and longitude coordinates, driving direction and driving speed, which is used to provide the cloud server and rescue platform with the precise geographical location of the collision event.

[0121] In some embodiments, the fusion judgment algorithm deployed locally on the vehicle may misjudge (e.g., misjudging non-collision events such as driving over speed bumps, driving over manhole covers, or road protrusions as collisions). Relying solely on local judgment results to directly trigger subsequent actions may lead to unnecessary braking, affecting vehicle operating efficiency and even causing traffic congestion. By sending road collision information to a cloud server, where a cloud-based safety officer performs a secondary manual confirmation in conjunction with external video footage, it is possible to effectively distinguish between real collisions and false alarms. In the event of a real collision, the rescue process can be initiated promptly, and in the event of a false alarm, a remote recovery command can be issued to quickly restore the vehicle to normal operation, avoiding operational interruptions caused by mishandling and achieving complete closed-loop management of collision event handling.

[0122] In some embodiments, when the target collision parameters are greater than the collision threshold, the vehicle determines that a road collision event has occurred. While performing a braking operation locally, it uploads the external video, location information and target collision parameters to the cloud, where a cloud safety officer performs a second manual confirmation: if it is a false alarm, a recovery command is issued; if it is a real collision, a rescue command is issued, thus achieving a closed-loop handling of vehicle-cloud collaboration.

[0123] Through the above methods, while the vehicle completes automatic collision event detection and emergency braking locally, the vehicle-cloud collaborative confirmation mechanism enables remote classification and closed-loop handling of collision events, effectively avoiding the problems of false alarms leading to operational interruptions or the inability to provide timely assistance in the event of a real collision.

[0124] In this embodiment, by introducing an independent physical collision detection link and performing multi-source fusion judgment with environmental perception information and passenger status information, low-speed minor scratches or low-visibility collisions can be accurately identified. This effectively avoids the problem of missed judgments and false judgments caused by the failure of a single environmental perception link in scenarios with limited perception, such as severe weather or sensor obstruction. Combined with the vehicle-cloud collaborative confirmation mechanism, the vehicle can drive based on control commands issued by the cloud, realizing remote classification and closed-loop handling of collision events. This improves the reliability of collision detection and ensures the operational safety of fully autonomous vehicles.

[0125] Figure 4 This is a schematic diagram illustrating a possible process for determining target collision parameters, provided in an embodiment of this application. Please refer to... Figure 4 The method may include:

[0126] S401. Align the first environment information, the first state information, and the first collision information to obtain the second environment information, the second state information, and the second collision information.

[0127] The second environmental information can refer to the environmental perception data obtained under a unified reference time and a unified vehicle coordinate system after the first environmental information is timestamped and mapped to a coordinate system. It includes information such as obstacle type, location, speed and direction of movement.

[0128] The second state information can refer to the passenger state data obtained under a unified reference time and a unified vehicle coordinate system after the first state information is timestamped and mapped to a coordinate system. It includes information such as passenger seating position, changes in human posture and displacement status.

[0129] The second collision information can refer to the collision detection data obtained under a unified reference time and a unified vehicle coordinate system after aligning the first collision information with timestamps and mapping the coordinate system. It includes the collision occurrence sign and the corresponding vehicle collision position information.

[0130] In some embodiments, the second environmental information, the second state information, and the second collision information can be obtained by: aligning the first environmental information, the first state information, and the first collision information with timestamps to obtain multi-source perception information at the same time; mapping the multi-source perception information at the same time to the same vehicle coordinate system to generate the second environmental information, the second state information, and the second collision information.

[0131] In some embodiments, the sampling frequencies and clock sources of the various vehicle sensors (environmental perception sensors, OMS image acquisition devices, and collision detection sensors) may differ, and directly using the raw data for fusion would lead to temporal mismatch. By obtaining the timestamp information recorded in the data packets of each sensor, and using a unified time base as a reference, interpolation or nearest neighbor matching methods are employed to synchronize the data collected by each sensor to the same moment, obtaining multi-source perception information at the same moment, and ensuring the temporal consistency of various types of information in subsequent fusion determination.

[0132] In some embodiments, the data collected by each sensor resides in different local coordinate systems (e.g., the camera is in the pixel coordinate system, the lidar is in the radar coordinate system, and the collision detection sensor is in the vehicle body area number coordinate system), making direct spatial correlation impossible. By pre-calibrating the extrinsic parameter matrix (including the rotation matrix R and the translation vector t) of each sensor relative to the rear axle center (or the vehicle's center of mass), the spatial position information in the data collected by each sensor is transformed to the same vehicle coordinate system through the extrinsic parameter matrix, ensuring the spatial consistency of various types of information in subsequent fusion determination.

[0133] The above method can align multi-source sensing information with inconsistent time and space benchmarks to the same time and coordinate system, ensuring the consistency of various types of information in time and space in subsequent fusion and determination, and providing a reliable data foundation for accurately determining target collision parameters.

[0134] S402. Determine the target collision parameters based on the second environmental information, the second state information, and the second collision information.

[0135] In some embodiments, a neural network model can automatically learn the complex interaction relationships and optimal fusion weights among the three types of information, avoiding subjective biases caused by manually designing fusion rules and weight allocations, and adaptively capturing the contribution of each information source to the collision determination result under different perception scenarios.

[0136] In some embodiments, the target collision parameters can be determined based on the second environmental information, the second state information, and the second collision information in the following manner: the second environmental information, the second state information, and the second collision information are input into the prediction model to predict the target collision parameters.

[0137] A prediction model can refer to a neural network model used for end-to-end fusion and determination of multi-source sensing information. Its input is a spatiotemporally aligned tensor of multi-source sensing information, and its output is the target collision parameters.

[0138] For example, the prediction model can employ a multimodal fusion network structure based on an attention mechanism. First, corresponding feature extraction branches are designed for the three types of input information: the environment perception branch uses a feature encoding network to extract features from the LiDAR point cloud and combines it with a 2D image semantic feature encoding network to encode semantic features from the surround view image, outputting an environment feature vector; the passenger state branch uses a lightweight pose estimation network to detect human keypoints in the OMS image, extracting passenger position and pose feature vectors; the collision detection branch encodes the collision signal intensity and collision location region collected by the collision sensor into a collision event feature vector. Then, the feature vectors output from the three branches are concatenated along the channel dimension, and an adaptive weight allocation is performed on the feature channels of different modalities through a cross-modal attention layer to obtain multimodal fusion features. Finally, the fusion features are input into a classification network composed of stacked fully connected layers, and the target collision parameters are output after mapping by an activation function.

[0139] The training process of the prediction model is as follows: The training dataset comes from real-vehicle road data and collision / non-collision scenario data generated by the simulation platform, including time-series data of multi-source perception information under various weather conditions (sunny, rainy, snowy, foggy), different lighting conditions (daytime, nighttime, backlight), and various perception-limited scenarios. The data annotation process adopts manual annotation before and after the collision event, and the collision ground truth label is marked for each sample at each time point according to the actual collision situation (0 indicates no collision, 1 indicates a collision).

[0140] During training, stochastic gradient descent or its variant optimizer is used to minimize the loss function. The weight parameters of each layer of the network are updated through backpropagation, so that the target collision parameters output by the model continuously approach the true label, and finally a well-trained prediction model is obtained.

[0141] By employing the above method, the complex mapping relationship between three types of multi-source perception information and collision events is learned end-to-end through a neural network model. This enables the adaptive capture of the contribution of each information source to the collision determination result under different perception scenarios, thereby improving the accuracy of target collision parameter determination and scene generalization ability.

[0142] Optionally, the second environmental information, the second state information, and the second collision information can be input into the prediction model to output the environmental perception confidence, passenger state confidence, and physical collision confidence. The three confidences are then weighted and summed to obtain the target collision parameters.

[0143] Specifically, the at most-output prediction model is input into the second environmental information, second state information, and second collision information. This model includes a feature extraction network and three parallel output branches. The feature extraction network fuses and encodes the three types of information into a multimodal feature representation. The three parallel branches map this shared feature and independently output corresponding confidence scores. Specifically, the first branch outputs the environmental perception confidence score, used to determine the degree of certainty that a collision has occurred based on environmental perception information; the second branch outputs the passenger state confidence score, used to determine the degree of certainty that a collision has occurred based on passenger state information; and the third branch outputs the physical collision confidence score, used to determine the degree of certainty that a collision has occurred based on physical collision information. Higher values ​​indicate a higher degree of confidence from which the information source believes a collision has occurred. After joint training, once the model converges, it can be used to simultaneously calculate the three confidence scores during the online inference phase.

[0144] After acquiring the environmental perception confidence level, passenger state confidence level, and physical collision confidence level, corresponding fusion weights are assigned to each of the three confidence levels. The weight assignment is determined based on two aspects: the prior reliability value of each information source in the current perception scenario and the dynamic confidence level.

[0145] Among them, the reliability prior value is obtained by pre-calibration based on the inherent performance parameters of each sensor (e.g., the point cloud quality degradation coefficient of lidar in rainy and foggy weather, and the signal stability coefficient of collision detection sensor under all-weather conditions), reflecting the inherent reliability of the sensor itself; the dynamic confidence is dynamically adjusted according to the real-time data quality at the current moment (e.g., the target detection box confidence score of the environmental perception module, the human key point detection confidence score of the OMS module, and the signal-to-noise ratio of the collision detection sensor).

[0146] The final fusion weights of each information source at the current moment are determined by combining the two factors: environmental perception fusion weight, passenger state fusion weight, and physical collision fusion weight. Finally, the three confidence levels are weighted and summed according to the following formula to obtain the target collision parameters.

[0147] By using a dynamic weight allocation mechanism, the target collision parameters can adaptively depend on the most reliable information source at the current moment under different perception scenarios, thereby improving the accuracy and robustness of collision determination.

[0148] In this embodiment, by aligning the first environmental information, the first state information, and the first collision information with timestamps and mapping them to a coordinate system, multi-source perception information with a unified spatial reference at the same time is obtained, and the target collision parameters are determined based on the aforementioned multi-source perception information. This method can accurately identify low-speed minor scratches or low-visibility collisions of vehicles in perception-limited scenarios such as severe weather or sensor obstruction, based on the weighted fusion of multi-source perception information. It achieves the unification of heterogeneous perception data in both time and space dimensions, eliminates deviations caused by sampling time differences between different sensors and inconsistent coordinate references, avoids the problem of missed or false judgments caused by the easy failure of single environmental perception in perception-limited scenarios, and improves the reliability of collision detection for fully autonomous vehicles.

[0149] Figure 5 This is a schematic diagram illustrating a possible process for determining target collision parameters, provided in an embodiment of this application. Please refer to... Figure 5 The method may include:

[0150] S501. Determine the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information.

[0151] The first collision parameter can refer to the confidence score output after the first environmental information is input into the first model, which is used to represent the confidence score of the environmental perception information in determining that a vehicle collision event has occurred.

[0152] The second collision parameter can refer to the confidence score output after the first state information is input into the second model, which is used to represent the passenger state information to determine that a vehicle collision event has occurred.

[0153] The third collision parameter can refer to the confidence score output by the third model after the first collision information is input into the third model, which is used to represent the physical collision information to determine that a vehicle has collided.

[0154] In some embodiments, the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information can be determined in the following manner: the first environmental information is input into the first model to obtain the first collision parameter; the first state information is input into the second model to obtain the second collision parameter; and the first collision information is input into the third model to obtain the third collision parameter.

[0155] The first model can refer to a neural network model used to determine the collision confidence of environmental perception information. Its input is environmental perception data, and its output is the environmental perception collision confidence, i.e., the first collision parameter.

[0156] The second model can refer to a neural network model that determines the collision confidence level based on passenger status information. Its input is passenger status data, and its output is the passenger status collision confidence level, i.e., the second collision parameter.

[0157] The third model can refer to a neural network model that determines the collision confidence level of physical collision information. Its input is collision detection data, and its output is the physical collision confidence level, i.e., the third collision parameter.

[0158] In some embodiments, the first, second, and third models can employ the same network architecture, for example, a binary classification structure based on a multi-layer fully connected network. The only difference between the models is that the input layer dimension is adapted to the feature dimension of their respective input data, and the output layer is mapped to the 0-1 interval using the Sigmoid function to output the corresponding collision confidence. The three models are trained independently, with parameters optimized using datasets of their respective modalities. After training, they are used in the online inference stage.

[0159] Specifically, the first model receives first environmental information as input. This first environmental information includes data collected by environmental perception sensors, such as obstacle type, location, speed, and direction of movement. This multi-source heterogeneous data is first preprocessed into a feature vector of uniform dimension, and then input into the first model. The first model performs layer-by-layer nonlinear transformation and dimensionality reduction on the input features through a multi-layer fully connected network, extracting deep semantic features related to collision determination. Finally, the output layer is mapped to the 0-1 interval via a Sigmoid function to obtain the first collision parameters.

[0160] The second model receives the first state information as input. This first state information includes passenger seating position, human body keypoint coordinates, and vehicle motion state data (e.g., longitudinal acceleration and lateral acceleration) collected by the OMS system. This data is first concatenated into a fixed-dimensional feature vector, which is then input into the second model. The second model uses a multi-layer fully connected network to perform layer-by-layer nonlinear transformation and dimensionality reduction on the input features, extracting abnormal passenger posture and impact response features. Finally, the output layer is mapped to the 0-1 interval using a Sigmoid function to obtain the second collision parameters.

[0161] The third model receives the first collision information as input. The first collision information includes the signal strength collected by collision detection sensors in various regions of the vehicle body and the corresponding collision location region codes. The vehicle first arranges the signal strengths of each region into feature vectors according to a preset spatial order, and maps the collision location region codes into dense vectors through an embedding layer. The two are then concatenated and input into the third model. The third model performs layer-by-layer nonlinear transformations on the input features through a multi-layer fully connected network to extract the spatial distribution features of the collision signals. Finally, the output layer maps the signals to the 0-1 interval using a Sigmoid function to obtain the third collision parameters.

[0162] In some embodiments, the first model, the second model, and the third model are trained independently. The training datasets are derived from real-vehicle road data and collision / non-collision scenario data generated by the simulation platform. Each model is trained using data of its corresponding modality (environmental perception data, passenger state data, and physical collision data). The training datasets include time-series data under various weather conditions, different lighting conditions, and various perception-limited scenarios. The data annotation process for each model adopts manual annotation before and after the collision event. Based on the actual collision occurrence, a ground truth collision label is assigned to each sample at each time point (0 indicates no collision, 1 indicates a collision).

[0163] During training, stochastic gradient descent or its variant optimizers can be used to minimize the loss function. The weight parameters of each layer in the network are updated via backpropagation, allowing the collision confidence scores output by the model to continuously approach the true labels, ultimately resulting in well-trained models. During the inference phase, each model receives its own input information and calculates and outputs the first, second, and third collision parameters in real time.

[0164] In the above manner, the three models extract features and determine collision confidence for environmental perception information, passenger status information and physical collision information, respectively. Each model adopts the same network architecture but is trained independently. It can achieve optimal parameter convergence for the data distribution characteristics of its respective input modality, and provide reliable single-source confidence input for subsequent multi-source weighted fusion.

[0165] S502. Determine the target collision parameters based on the first collision parameters, the second collision parameters, and the third collision parameters.

[0166] In some embodiments, the target collision parameters can be determined based on the first collision parameters, the second collision parameters, and the third collision parameters in the following manner: determining the collision weights corresponding to each collision parameter; and calculating the target collision parameters based on the first collision parameters, the second collision parameters, the third collision parameters, and the corresponding collision weights.

[0167] Collision weights are used to indicate the proportion of collision parameters corresponding to each information source in the fusion calculation, reflecting the degree of contribution of each information source to the collision determination result in the current perception scenario.

[0168] In some embodiments, the current perception scene type can be determined based on environmental perception information. The prior reliability values ​​of each sensor differ under different scenes; collision detection sensors are less affected by weather, resulting in a relatively higher prior reliability value. Simultaneously, the real-time data quality of each information source is quantitatively determined based on the real-time state diagnostic information reported by each sensor (e.g., point cloud density, image sharpness, signal-to-noise ratio). The weights of each information source are determined by combining the prior reliability values ​​and real-time data quality of each information source in the current scene, with all three satisfying normalization constraints.

[0169] Optionally, a set of fixed base weights can be pre-calibrated based on the inherent reliability of each sensor under standard operating conditions. Based on this, the data quality deviation of each information source relative to the standard operating conditions is calculated according to the real-time status diagnostic information reported by each sensor at the current moment. This deviation is then mapped to a corresponding weight compensation amount; that is, positive compensation is applied when the real-time data quality is higher than the standard operating conditions, and negative compensation is applied when it is lower. The final fusion weight of each information source is the sum of the base weight and the compensation amount. After normalization, this satisfies the constraint that the sum of the three is 1, thus preserving the reasonable weight allocation under standard operating conditions while enabling adaptive adjustment based on real-time data quality.

[0170] In some embodiments, the final target collision parameters can be calculated by linearly weighting and summing each collision parameter with its corresponding weight. That is, the target collision parameters are obtained by weighted summing of the first collision parameter and its corresponding collision weight, the second collision parameter and its corresponding collision weight, and the third collision parameter and its corresponding collision weight.

[0171] By employing the above method, a dynamic weighting mechanism is used to weight and fuse the collision confidence of different information sources, enabling the target collision parameters to adaptively depend on the most reliable information source at the current moment under different perception scenarios. This improves the accuracy and robustness of collision determination for fully autonomous vehicles while enhancing the reliability of collision detection.

[0172] In this embodiment, the first environmental information, the first state information, and the first collision information are input into the corresponding first model, second model, and third model, respectively, to obtain their respective collision parameters. Then, based on the weights corresponding to each collision parameter, a weighted sum is performed to obtain the target collision parameter. This method, based on a collaborative mechanism of independent judgment from each information source and dynamic weight allocation, can accurately identify low-speed minor scratches or low-visibility collisions of vehicles in scenarios with limited perception, such as adverse weather or sensor obstruction. It avoids the problems of missed or false judgments caused by the easy failure of single environmental perception, thus improving the reliability of collision detection for fully autonomous vehicles.

[0173] Figure 6 This is a schematic diagram illustrating the process of determining vehicle control parameters as provided in an embodiment of this application. Please refer to... Figure 6 The method may include:

[0174] S601. When the target collision parameter is greater than the collision threshold, the vehicle control parameter is determined as the first control parameter for controlling the vehicle to perform deceleration and stopping operation.

[0175] A collision threshold can refer to a pre-set confidence threshold used to determine whether a vehicle has been involved in a road collision event. When the target collision parameters exceed this threshold, it is determined that a road collision event has occurred, and emergency response is required; when the target collision parameters do not exceed this threshold, it is determined that no road collision event has occurred, and normal driving continues.

[0176] The first control parameter can refer to the set of parameters generated when a road collision event is determined to have occurred, which is used to control the vehicle to perform deceleration and stop operations. It includes information such as target deceleration, braking pressure, and parking position, and is used to drive the vehicle braking system to execute the minimum risk strategy (MRM) to decelerate the vehicle to a stop.

[0177] In some embodiments, the target collision parameters are compared with a preset collision threshold. When the target collision parameters are greater than the collision threshold, it indicates a high degree of confidence in the occurrence of a collision event. The vehicle determines that a road collision event has occurred and immediately triggers the Minimum Risk Management (MRM) strategy, generating first control parameters and sending them to the braking system. Based on the target deceleration and braking pressure in the first control parameters, the braking system controls the vehicle to perform deceleration, pull over, and stop safely, while simultaneously activating hazard warning lights to prevent secondary collisions.

[0178] In this way, when a collision is detected, the vehicle can immediately trigger the MRM strategy to control the vehicle to stop within the shortest safe distance, effectively preventing the consequences of the collision.

[0179] S602. When the target collision parameter is less than or equal to the collision threshold, the vehicle control parameter is determined to be a second control parameter used to control the vehicle to maintain its current motion state.

[0180] The second control parameter can refer to the set of parameters generated when it is determined that no road collision event has occurred, which are used to control the vehicle to maintain its current state of motion. It includes the target driving torque to maintain the current vehicle speed and the target steering angle to maintain the current driving direction, and is used to drive the vehicle's power system and steering system to maintain normal operation.

[0181] In some embodiments, when the target collision parameter is less than or equal to the collision threshold, it indicates that the credibility of the current collision event is low, and the vehicle determines that no road collision event has occurred. At this time, the vehicle generates a second control parameter and sends it to the power system and steering system to control the vehicle to maintain its current motion state and continue driving, while continuously acquiring multi-source perception information for the next moment and continuing to perform collision detection.

[0182] By using the above method, when it is determined that no collision has occurred, the vehicle continues to drive in its current state of motion, effectively avoiding unnecessary braking and operational interruptions caused by misjudgment.

[0183] In this embodiment, by comparing the target collision parameters with a collision threshold and generating corresponding vehicle control parameters based on the comparison result, continuous confidence levels are converted into discrete driving control commands. When the target collision parameters are greater than the collision threshold, the MRM strategy is triggered to control the vehicle to perform a deceleration and stop operation; when the target collision parameters are less than or equal to the collision threshold, the vehicle is controlled to maintain its current motion state, avoiding unnecessary braking and operational interruptions caused by misjudgment. The above method can ensure the operational safety of fully autonomous vehicles while also considering operational efficiency.

[0184] Figure 7 This is a schematic flowchart illustrating another vehicle collision detection method provided in an embodiment of this application. Please refer to... Figure 7 As shown, the method may include the following steps:

[0185] S701, Receives road collision information sent by the vehicle.

[0186] In some embodiments, after receiving road collision information sent by the vehicle, the following steps can be taken: the road collision information is verified to obtain a verification result; if the verification result is a failure, the verification result is sent to the vehicle so that the vehicle can perform a data retransmission operation.

[0187] The verification result can refer to the judgment result of the cloud server after performing integrity verification on the received road collision information data packet. The verification result can be either verification passed or verification failed.

[0188] In some embodiments, after receiving road collision information uploaded by the vehicle, the cloud server first performs an integrity check on the data packet to verify whether the data was lost or damaged during transmission. If the check fails, it indicates that the data packet is lost or damaged. The cloud server then sends a retransmission request to the vehicle. Upon receiving the retransmission request, the vehicle re-uploads the road collision information until the check passes or the number of retransmissions reaches a preset limit. This method ensures the integrity and reliability of data transmission between the vehicle and the cloud.

[0189] In some embodiments, if the verification result is successful, it indicates that the cloud server obtained the verification result during the verification process as the data packet is complete and error-free. The cloud server confirms successful reception and can continue to execute S702.

[0190] By employing the above methods and data integrity verification and retransmission mechanisms, we ensure that the road collision information uploaded by the vehicle is not lost or tampered with during transmission, thus providing a reliable data foundation for subsequent remote confirmation.

[0191] S702. Determine the vehicle's collision control strategy based on road collision information.

[0192] Among them, the collision control strategy is used to control the vehicle to perform corresponding driving operations.

[0193] In some embodiments, a collision control strategy for a vehicle may be determined based on road collision information in the following manner: when it is determined based on road collision information that the vehicle has not been involved in a road collision event, the collision control strategy is determined to be a recovery strategy for controlling the vehicle to resume normal driving.

[0194] Recovery strategy refers to the control strategy generated by the cloud-based safety operator after reviewing the external video uploaded by the vehicle and determining that the vehicle did not actually collide (i.e., it was a false alarm), which is used to control the vehicle to exit the emergency response state and resume normal driving.

[0195] In some embodiments, after the cloud server receives and successfully verifies the road collision information uploaded by the vehicle, it triggers the scheduling of the remote monitoring platform and generates a confirmation work order in the safety operator's console. The safety operator manually reconfirms the collision event by reviewing the vehicle's external surround view video uploaded by the vehicle, combined with the target collision parameters and vehicle position information. If the safety operator confirms through video that there are no collision marks or objects around the vehicle, and determines that the current collision event is a false alarm (e.g., a false trigger caused by a non-collision event such as the vehicle driving over a speed bump, driving over a manhole cover, or a road protrusion), then the collision control strategy is determined to be a recovery strategy. The cloud server generates a corresponding control command based on the recovery strategy, the command content of which is to allow the vehicle to resume normal driving, so that the vehicle exits the minimum risk strategy (MRM) state and continues to perform operational tasks.

[0196] By employing the above methods, in the event of a false alarm, the vehicle can quickly resume normal operation through manual confirmation by a cloud-based safety officer and remote recovery commands. This avoids unnecessary parking and operational interruptions caused by misjudgments, effectively ensuring the operational efficiency of fully driverless vehicles.

[0197] In some embodiments, a vehicle collision control strategy can be determined based on road collision information in the following manner: when it is determined that a road collision event has occurred based on road collision information, the collision control strategy is determined to be a handling strategy for pushing road collision information to the platform.

[0198] The response strategy can refer to the control strategy generated by the cloud-based safety operator after reviewing the external video uploaded by the vehicle and determining that a road collision has indeed occurred. This strategy is used to push accident information to the rescue platform and initiate the emergency response process.

[0199] In some embodiments, after the cloud server receives and successfully verifies the road collision information uploaded by the vehicle, it triggers the dispatch of the remote monitoring platform and generates a confirmation work order in the safety officer's interface. The safety officer manually confirms the collision event by viewing the external surround view video uploaded by the vehicle, combined with the target collision parameters and vehicle position information. If the safety officer confirms through the video that a real collision has occurred and determines that the current event is a real collision, then the collision control strategy is determined to be the handling strategy. The cloud server generates corresponding control instructions based on the handling strategy. On the one hand, it pushes the road collision information, including vehicle position information, collision event data, and external video, to the rescue platform, automatically triggering the rescue process; on the other hand, it sends rescue handling instructions to the vehicle. After receiving the instructions, the vehicle initiates the emergency comfort and interaction program for passengers inside the vehicle and waits for the arrival of rescue personnel.

[0200] Through the above methods, in real collision scenarios, the rescue process can be initiated in a timely manner by the manual confirmation of the safety officer in the cloud and the rescue and disposal instructions can be issued to the vehicle, ensuring that the accident is responded to quickly and handled properly, and maximizing the safety of personnel.

[0201] S703. Based on the collision control strategy, generate control commands and issue control commands to the vehicle so that the vehicle can drive based on the control commands.

[0202] In some embodiments, the cloud server generates corresponding control commands based on the collision control strategy determined in S702. If the collision control strategy is a recovery strategy, the cloud server generates a recovery operation command, which allows the vehicle to resume normal driving and exit the minimum risk strategy state, and sends it to the vehicle via the vehicle-cloud communication link. If the collision control strategy is a handling strategy, the cloud server generates a rescue handling command, which instructs the vehicle to remain parked and activate the emergency passenger reassurance program, and sends it to the vehicle via the vehicle-cloud communication link, while simultaneously pushing road collision information to the rescue platform. The vehicle receives and executes the control commands sent from the cloud, completing the closed-loop handling of the collision event.

[0203] In this embodiment, the cloud server receives road collision information reported by the vehicle, which is then manually confirmed by a safety officer using external video footage. Based on the confirmation results, the collision control strategy for the vehicle is determined, and corresponding control commands are generated and sent to the vehicle. This enables remote classification and closed-loop handling of collision events, allowing for timely initiation of rescue procedures in the event of a real collision and remote restoration of vehicle operation in the event of a false alarm. This effectively improves the reliability and operational continuity of handling collision events in fully autonomous vehicles.

[0204] Below, in conjunction with Figure 8 The system architecture of this application will be described in detail.

[0205] Figure 8 This is a schematic diagram of the system architecture provided for an embodiment of this application. Please refer to [link / reference]. Figure 8 The system architecture consists of four main modules: a vehicle-side collision detection system, an autonomous driving system, a gateway and vehicle-to-cloud communication module, and a remote monitoring and intervention system. These modules communicate bidirectionally via pre-defined links to collaboratively detect, handle, and confirm collision events.

[0206] The vehicle-side collision detection system consists of collision detection sensors and a collision detection controller deployed in various areas of the vehicle body. The collision detection sensors collect collision signals from various locations on the vehicle body in real time, and the collision detection controller performs signal preprocessing on the raw signals output by the sensors, outputting the collision detection results and corresponding collision location information, providing independent physical collision evidence for subsequent fusion and judgment.

[0207] The autonomous driving system, serving as the core of onboard decision-making, comprises an environmental perception sensor array, an OMS (On-Site Management System) passenger detection system, a fusion algorithm module, and an MRM (Mount Response Management) module. The environmental perception sensor array collects environmental perception data around the vehicle using cameras, millimeter-wave radar, and lidar. The OMS passenger detection system collects passenger status data using in-vehicle image acquisition equipment and an accelerometer. The fusion algorithm module receives the aforementioned environmental perception data, passenger status data, and collision detection results and location information output by the onboard collision detection system. It performs spatiotemporal alignment and confidence fusion of multi-source information to generate a collision event determination result. When the fusion determines a real collision, the MRM module immediately sends a braking command to the vehicle's braking system, controlling the vehicle to decelerate, pull over, and safely stop.

[0208] The gateway and vehicle-to-cloud communication module comprises a gateway controller and a 5G vehicle-mounted communication terminal (T-box). Its core function is to enable low-latency bidirectional communication between the vehicle and the cloud server. This module uploads collision event data, external video, and vehicle location information to the cloud in real time, while simultaneously receiving control commands from the cloud server. Furthermore, the module employs encryption and verification mechanisms to ensure the security and integrity of data transmission and features offline backup functionality. In extreme scenarios, local caching and offline retransmission mechanisms prevent communication failures that could lead to a loss of control.

[0209] The remote monitoring and intervention system consists of a remote cloud platform and a safety operator's workstation. After receiving road collision information uploaded from the vehicle, the system triggers a safety operator's order on the remote monitoring platform. The safety operator then manually confirms the incident by reviewing the external video uploaded from the vehicle. If the incident is determined to be a false alarm, the safety operator issues a remote resumption command, and the vehicle exits MRM mode and resumes normal operation. If the incident is determined to be a real collision, the safety operator issues a rescue command, and the vehicle automatically pushes accident information to the rescue platform and initiates passenger emergency reassurance and interaction while awaiting the arrival of rescue personnel.

[0210] Through the above system architecture, the vehicle subsystem and the cloud subsystem communicate bidirectionally through a low-latency interactive link, forming a closed-loop security architecture that includes multi-source detection at the vehicle end, vehicle-mounted fusion judgment, local emergency response, vehicle-cloud collaborative confirmation, and full-process traceability.

[0211] Below, in conjunction with Figure 9 The process of vehicle collision detection is explained in detail.

[0212] Figure 9 This is a schematic flowchart illustrating another vehicle collision detection method provided in this application. Please refer to... Figure 9The method may include: the process begins by real-time acquisition of collision and scratch signals at various locations on the vehicle body using collision detection sensors (e.g., elastic wave sensors, accelerometers, or barometric pressure sensors) deployed in various areas of the vehicle body, especially for comprehensive detection of minor scratches at low speeds. The sensor signals are preprocessed by a dedicated controller to obtain the first collision information and the corresponding collision location information. Simultaneously, environmental perception data around the vehicle is acquired through an environmental perception sensor group, and passenger status data is acquired through an OMS passenger detection system. The aforementioned multi-source perception information is uniformly fed into the autonomous driving domain controller. Subsequently, the autonomous driving fusion algorithm module receives the detection results from the vehicle-side collision detection system, combines the environmental perception data and the OMS passenger detection results, performs spatiotemporal alignment and confidence-weighted fusion of the multi-source information, and generates a collision event determination result, thereby improving the reliability of the collision detection results and reducing the false alarm and false negative rates. When the collision is determined to be a real collision, the local emergency response module immediately triggers the minimum risk strategy (MRM), sending a braking command to the vehicle's braking system to execute deceleration, pull over, and safely stop the vehicle to prevent further escalation. Simultaneously, the gateway and vehicle-to-cloud communication module upload collision event data, location information, and external video footage to the cloud in real time. After receiving the collision event information uploaded from the vehicle, the remote monitoring and intervention system performs a secondary manual verification by reviewing the external video footage to distinguish between a real collision and a false alarm: if determined to be a false alarm, the safety officer issues a remote resumption command, and the vehicle exits MRM mode and resumes normal operation; if determined to be a real collision, the safety officer issues a rescue command, and the vehicle automatically pushes accident information to the rescue platform and initiates passenger emergency reassurance and interaction.

[0213] In this embodiment, by fusing collision detection sensors deployed across the vehicle body with autonomous driving environmental perception data, a dual evidence chain of physical perception and environmental perception is formed. This effectively identifies scenarios with limited perception, long-tail scenarios, and minor low-speed scratches, achieving highly reliable, no-miss detection of collision events and significantly reducing false alarm rates. After a collision determination, the MRM strategy triggers a braking command within milliseconds, initiating braking within 50-100ms to effectively prevent secondary collisions. Furthermore, the combination of local automatic handling and remote manual confirmation ensures rapid response to collision events while allowing safety personnel to intervene and resolve algorithmic misjudgments. In false alarm scenarios, vehicle operation can be quickly restored; in real collision scenarios, rescue can be initiated promptly, forming a complete safety closed loop of detection, handling, confirmation, and rescue / recovery. Simultaneously, all data throughout the collision event process, including perception data, fusion determination results, handling commands, external video, and manual confirmation records, is uploaded to cloud storage. This data can be used for collision event tracing and liability determination, providing data support for fusion algorithm optimization and sensor deployment optimization, continuously improving system reliability. The above mechanism enables vehicles to quickly resume operation without on-site human intervention in false alarm scenarios, effectively reducing the impact of mishandling on the operational efficiency of autonomous vehicles, ensuring safety while taking into account operational continuity, and providing technical support for the implementation of fully autonomous driving.

[0214] Figure 10 This is a schematic diagram of a vehicle collision detection device provided in an embodiment of this application. Please refer to... Figure 10 The vehicle collision detection device 1000 includes: an acquisition module 1001, a first determination module 1002, a second determination module 1003, and a transmission module 1004, wherein:

[0215] The acquisition module 1001 is used to acquire multi-source perception information of the vehicle during driving, including first environmental information, first passenger state information and first collision information.

[0216] The first determining module 1002 is used to determine the target collision parameters based on the first environmental information, the first state information and the first collision information. The target collision parameters are used to determine whether the vehicle has caused a road collision event.

[0217] The second determining module 1003 is used to determine vehicle control parameters based on the target collision parameters. The vehicle control parameters are used to control the vehicle to perform corresponding driving operations.

[0218] The sending module 1004 is used to send the vehicle's road collision information to the cloud server when the target collision parameters indicate that a road collision event has occurred, so that the vehicle can drive based on the control commands issued by the cloud server.

[0219] In one possible implementation, the acquisition module 1001 is specifically used for:

[0220] Acquire collision signals collected by detection devices in various areas of the vehicle body;

[0221] The collision signal is preprocessed and the region is located to obtain the corresponding collision result and collision location information, and the collision result and collision location information are determined as the first collision information.

[0222] In one possible implementation, the first determining module 1002 is specifically used for:

[0223] Alignment processing is performed on the first environmental information, the first state information, and the first collision information to obtain the second environmental information, the second state information, and the second collision information;

[0224] Based on the second environmental information, the second state information, and the second collision information, the target collision parameters are determined.

[0225] In one possible implementation, the first determining module 1002 is specifically used for:

[0226] The first environmental information, the first state information, and the first collision information are time-stamped to obtain multi-source sensing information at the same moment.

[0227] Multi-source perception information at the same moment is mapped to the same vehicle coordinate system to generate second environmental information, second state information and second collision information.

[0228] In one possible implementation, the first determining module 1002 is specifically used for:

[0229] The second environmental information, the second state information, and the second collision information are input into the prediction model to predict the target collision parameters.

[0230] In one possible implementation, the first determining module 1002 is specifically used for:

[0231] Determine the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information;

[0232] The target collision parameters are determined based on the first collision parameters, the second collision parameters, and the third collision parameters.

[0233] In one possible implementation, the first determining module 1002 is specifically used for:

[0234] The first environmental information is input into the first model to obtain the first collision parameters;

[0235] The first state information is input into the second model to obtain the second collision parameters;

[0236] The first collision information is input into the third model to obtain the third collision parameters.

[0237] In one possible implementation, the first determining module 1002 is specifically used for:

[0238] Determine the collision weights corresponding to each collision parameter;

[0239] The target collision parameters are calculated based on the first collision parameter, the second collision parameter, the third collision parameter and the corresponding collision weights.

[0240] In one possible implementation, the second determining module 1003 is specifically used for:

[0241] When the target collision parameter is greater than the collision threshold, the vehicle control parameter is determined as the first control parameter used to control the vehicle to perform deceleration and stopping operations.

[0242] In one possible implementation, the second determining module 1003 is specifically used for:

[0243] When the target collision parameter is less than or equal to the collision threshold, the vehicle control parameter is determined as a second control parameter used to control the vehicle to maintain its current state of motion.

[0244] In one possible implementation, the sending module 1004 is specifically used for:

[0245] Acquire video and vehicle location information at the time of a road collision;

[0246] The video information, vehicle location information, and target collision parameters are sent to the cloud server as road collision information.

[0247] The vehicle collision detection device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be described again here.

[0248] Figure 11 This is a schematic diagram of another vehicle collision detection device provided in an embodiment of this application. Please refer to... Figure 11 The vehicle collision detection device 1100 includes: a receiving module 1101, a second determining module 1102, and a generating module 1103, wherein:

[0249] The receiving module 1101 is used to receive road collision information sent by the vehicle.

[0250] The third determining module 1102 is used to determine the vehicle's collision control strategy based on the road collision information. The collision control strategy is used to control the vehicle to perform corresponding driving operations.

[0251] The generation module 1103 is used to generate control commands according to the collision control strategy and issue control commands to the vehicle so that the vehicle can drive based on the control commands.

[0252] In one possible implementation, the second determining module 1102 is specifically used for:

[0253] When it is determined that no road collision event has occurred based on road collision information, the collision control strategy is determined to be a recovery strategy used to control the vehicle to resume normal driving.

[0254] In one possible implementation, the second determining module 1102 is specifically used for:

[0255] When a road collision event is determined based on road collision information, the collision control strategy is determined to be a handling strategy used to push road collision information to the platform.

[0256] In one possible implementation, the device further includes a processing module, which is specifically used for:

[0257] The road collision information is verified to obtain the verification results;

[0258] If the verification result is that the verification fails, the verification result is sent to the vehicle terminal so that the vehicle terminal can perform a data retransmission operation.

[0259] Figure 12 This is a schematic diagram of the structure of the vehicle collision detection device provided in an embodiment of this application. Figure 12 As shown, the vehicle collision detection device 1200 may include: a transceiver 1201, a processor 1202, and a memory 1203.

[0260] Processor 1202 executes computer execution instructions stored in memory, causing processor 1202 to perform the scheme in the above embodiments. Processor 1202 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0261] The memory 1203 is connected to the processor 1202 via the system bus and completes communication between them. The memory 1203 is used to store computer program instructions.

[0262] Transceiver 1201 can be used to obtain the task to be run and its configuration information.

[0263] The system bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0264] This application also provides a chip for executing instructions, which is used to execute the technical solution of the vehicle collision detection method in the above embodiments.

[0265] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the vehicle collision detection method described in the above embodiments.

[0266] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the vehicle collision detection method in the above embodiments.

[0267] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0268] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0269] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit integrating the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0270] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0271] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0272] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0273] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0274] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0275] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.

[0276] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0277] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A vehicle collision detection method, characterized in that, The method includes: Acquire multi-source perception information of the vehicle during its driving process, the multi-source perception information including first environmental information, first passenger state information and first collision information; Based on the first environmental information, the first state information, and the first collision information, target collision parameters are determined, and the target collision parameters are used to determine whether the vehicle has experienced a road collision event. Based on the target collision parameters, vehicle control parameters are determined, and the vehicle control parameters are used to control the vehicle to perform corresponding driving operations; When the target collision parameters indicate that the road collision event has occurred, the road collision information of the vehicle is sent to the cloud server so that the vehicle can drive based on the control commands issued by the cloud server.

2. The method according to claim 1, characterized in that, Obtaining the first collision information includes: Acquire collision signals collected by detection devices in various areas of the vehicle body; The collision signal is preprocessed and the region is located to obtain the corresponding collision result and collision location information, and the collision result and collision location information are determined as the first collision information.

3. The method according to claim 1, characterized in that, Determine the target collision parameters, including: The first environmental information, the first state information, and the first collision information are aligned to obtain the second environmental information, the second state information, and the second collision information. The target collision parameters are determined based on the second environmental information, the second state information, and the second collision information.

4. The method according to claim 3, characterized in that, Obtain the second environmental information, the second state information, and the second collision information, including: The first environmental information, the first state information, and the first collision information are time-stamped to obtain multi-source sensing information at the same moment. The multi-source perception information at the same moment is mapped to the same vehicle coordinate system to generate the second environmental information, the second state information, and the second collision information.

5. The method according to claim 3, characterized in that, Based on the second environmental information, the second state information, and the second collision information, the target collision parameters are determined, including: The second environmental information, the second state information, and the second collision information are input into the prediction model to predict the target collision parameters.

6. The method according to claim 1, characterized in that, Determine the target collision parameters, including: Determine the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information; The target collision parameters are determined based on the first collision parameters, the second collision parameters, and the third collision parameters.

7. The method according to claim 6, characterized in that, Determining the first collision parameter corresponding to the first environmental information, the second collision parameter corresponding to the first state information, and the third collision parameter corresponding to the first collision information includes: The first environmental information is input into the first model to obtain the first collision parameters; The first state information is input into the second model to obtain the second collision parameters; The first collision information is input into the third model to obtain the third collision parameters.

8. The method according to claim 6 or 7, characterized in that, Determining the target collision parameters based on the first collision parameters, the second collision parameters, and the third collision parameters includes: Determine the collision weights corresponding to each collision parameter; The target collision parameters are calculated based on the first collision parameters, the second collision parameters, the third collision parameters, and the corresponding collision weights.

9. The method according to claim 1, characterized in that, Based on the target collision parameters, vehicle control parameters are determined, including: When the target collision parameter is greater than the collision threshold, the vehicle control parameter is determined as the first control parameter for controlling the vehicle to perform a deceleration and stopping operation.

10. The method according to claim 9, characterized in that, Based on the target collision parameters, vehicle control parameters are determined, including: When the target collision parameter is less than or equal to the collision threshold, the vehicle control parameter is determined to be a second control parameter for controlling the vehicle to maintain its current motion state.

11. The method according to claim 1, characterized in that, Send the road collision information of the vehicle to the cloud server, including: Acquire video information and vehicle location information at the time of the road collision event; The video information, the vehicle location information, and the target collision parameters are sent to the cloud server as the road collision information.

12. A vehicle collision detection method, characterized in that, The method includes: Receive road collision information sent from the vehicle; Based on the road collision information, a collision control strategy for the vehicle is determined, and the collision control strategy is used to control the vehicle to perform corresponding driving operations; According to the collision control strategy, a control command is generated and issued to the vehicle so that the vehicle can drive based on the control command.

13. The method according to claim 12, characterized in that, Based on the road collision information, a collision control strategy for the vehicle is determined, including: When it is determined based on the road collision information that the vehicle has not been involved in a road collision event, the collision control strategy is determined to be a recovery strategy for controlling the vehicle to resume normal driving.

14. The method according to claim 12 or 13, characterized in that, Based on the road collision information, a collision control strategy for the vehicle is determined, including: When it is determined that the vehicle has experienced a road collision event based on the road collision information, the collision control strategy is determined to be a handling strategy for pushing the road collision information to the platform.

15. The method according to claim 12, characterized in that, After receiving the road collision information sent by the vehicle, the method further includes: The road collision information is verified to obtain the verification result; When the verification result is that the verification fails, the verification result is sent to the vehicle terminal so that the vehicle terminal can perform a data retransmission operation.

16. A vehicle collision detection device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-15.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-15.

18. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1-15.

Images