Vehicle safety control method, vehicle and electronic equipment

Abstract

The invention relates to a vehicle safety control method, a vehicle and electronic equipment, and belongs to the technical field of vehicle safety, and the method comprises the steps: obtaining front multi-source sensing information, and judging whether a target pedestrian has a road crossing intention through a multi-mode intention prediction model based on the front multi-source sensing information; when it is determined that the target pedestrian has the intention of crossing the road, performing lateral risk identification based on the acquired lateral target motion data and pedestrian motion data to obtain a lateral risk level; performing rear risk identification based on the obtained rear target motion data to obtain a rear risk level; determining a comprehensive risk level based on a preset mapping relation according to the lateral risk level and the rear risk level; and according to the comprehensive risk level, performing control configuration on at least one of the vehicle exterior voice module, the vehicle exterior light module and the vehicle exterior interaction screen, and executing a corresponding vehicle exterior interaction control strategy.

Description

Technical Field

[0001] This application relates to the technical field of vehicle safety, and more particularly to a vehicle safety control method, a vehicle, and electronic equipment. Background Technology

[0002] In urban traffic environments, yielding to pedestrians is an important aspect of civilized driving. However, the act of slowing down or stopping at crosswalks to yield to pedestrians can itself create new safety hazards. On the one hand, a vehicle's sudden slowing down to yield can easily lead to a rear-end collision as following vehicles are unable to react in time. On the other hand, when pedestrians are crossing the road, there may be oncoming vehicles in the blind spots of both the driver and the pedestrian, greatly increasing the risk of collisions.

[0003] To improve pedestrian safety when crossing the street, various active safety solutions based on vehicle perception and early warning have been proposed in existing technologies. For example, by using onboard sensors or vehicle-to-infrastructure (V2I) technology to detect pedestrians in front or in blind spots, when a collision risk is determined, the system will warn or intervene to the driver through in-vehicle indicator lights, audible warnings, or automatic emergency braking to avoid a collision between the vehicle and the pedestrian.

[0004] However, existing solutions primarily focus on addressing the risk of direct collisions between vehicles and pedestrians. Their warnings and interventions are mainly geared towards the driver, aiming to bring the vehicle to a stop to avoid an accident. These solutions lack multi-directional, coordinated risk perception and interactive warnings for the vehicle, pedestrians, vehicles behind, and vehicles to the side in pedestrian-yielding scenarios. Therefore, they cannot effectively prevent secondary accidents or risks that may arise from yielding behavior, and their warning dimensions are relatively singular, lacking sufficient interactive capabilities. Summary of the Invention

[0005] This application addresses, to at least some extent, one of the technical problems in the related art.

[0006] Therefore, this application aims to provide a vehicle safety control method, a vehicle, and electronic equipment.

[0007] To achieve the above objectives, in a first aspect, this application provides a vehicle safety control method, comprising: Acquire multi-source perception information ahead, and based on the multi-source perception information ahead, use a multimodal intent prediction model to determine whether the target pedestrian intends to cross the road; When it is determined that the target pedestrian has the intention to cross the road, lateral risk identification is performed based on the acquired lateral target movement data and pedestrian movement data to obtain the lateral risk level; and rear risk identification is performed based on the acquired rear target movement data to obtain the rear risk level. Based on the lateral risk level and the rear risk level, a comprehensive risk level is determined according to a preset mapping relationship; Based on the comprehensive risk level, control configuration is applied to at least one of the vehicle exterior voice module, vehicle exterior lighting module, and vehicle exterior interactive screen, and the corresponding vehicle exterior interactive control strategy is executed.

[0008] The technical solution incorporates a multimodal intent prediction model and a joint risk level assessment mechanism to accurately identify pedestrian crossing intentions and dynamic risks around vehicles during yielding scenarios. Upon detecting a pedestrian's intention to cross, the system further combines lateral and rear risk levels to determine a comprehensive risk level, which then drives the external voice module, external lighting module, and external interactive screen for proactive interactive control. Through this interactive response to external traffic participants, vehicles can proactively communicate their status and intentions to pedestrians or surrounding vehicles within a controllable risk range, effectively enhancing external early warning capabilities and traffic coordination efficiency, and significantly improving traffic safety and interactive intelligence during yielding processes.

[0009] In some embodiments of this application, lateral risk identification is performed based on acquired lateral target motion data and pedestrian motion data, and the resulting lateral risk levels include: Acquire the first motion state information of the lateral moving target and the second motion state information of the pedestrian; Based on the first motion state information and the second motion state information, the first motion trajectory of the lateral moving target and the second motion trajectory of the pedestrian are predicted respectively. Determine whether there is a conflict point between the first motion trajectory and the second motion trajectory; If a conflict point exists, the lateral estimated collision time between the moving target and the pedestrian is calculated based on the conflict point, and the lateral risk level is determined based on the lateral estimated collision time.

[0010] In this technical solution, trajectory prediction is performed separately based on the motion state information of the lateral moving target and the pedestrian. By determining whether there are conflict points between the trajectories, the estimated lateral collision time is calculated, enabling proactive identification of potential intersections and collision risks. Compared to traditional methods that rely solely on static parameters such as distance or speed for risk assessment, the introduction of dynamic trajectory prediction and a time-dimensional collision prediction mechanism allows for more accurate identification of the likelihood of lateral traffic conflicts. By quantifying the estimated lateral collision time into a risk level, the system can more rationally drive external interaction responses, thereby issuing early warnings to pedestrians or oncoming vehicles, enhancing external safety perception capabilities and proactive warning levels, and effectively reducing the risk of lateral collision accidents.

[0011] In some embodiments of this application, determining whether there is a conflict point between the first motion trajectory and the second motion trajectory includes: If there is no conflict point, calculate the shortest distance between the first motion trajectory and the second motion trajectory within the prediction time window; The lateral risk level is determined based on the relationship between the shortest distance and the preset safe distance threshold.

[0012] In this technical solution, when there is no clear point of conflict between a lateral moving target and the predicted trajectory of a pedestrian, the shortest distance within the prediction time window is calculated and compared with a preset safe distance threshold to achieve a quantitative assessment of potential approach risks. This solution avoids missing risks due to non-intersecting trajectories and enhances the system's robustness in risk identification under complex traffic environments. By incorporating the shortest distance into the criterion for determining lateral risk levels, the system can respond in advance before spatial intersections occur, ensuring sufficient warning time for vehicles during external interactions, thereby improving lateral safety protection capabilities and the accuracy and reliability of the interaction system.

[0013] In some embodiments of this application, calculating the estimated lateral collision time between the laterally moving target and the pedestrian based on the collision point includes: Based on the conflict point, calculate the first time when the pedestrian arrives at the conflict point; Calculate the second time when the lateral moving target arrives at the point of conflict; The smaller value between the first time and the second time is determined as the expected lateral collision time; The lateral risk level is determined based on the relationship between the estimated lateral collision time and a preset first time threshold.

[0014] In this technical solution, by introducing a conflict point and calculating the arrival time of pedestrians and lateral moving targets at that point, the smaller value is used as the predicted lateral collision time. This effectively reflects the earliest possible moment of risk of collision, thus enabling early identification of imminent risks. Compared to traditional static judgment methods based on distance or speed, this approach has higher dynamic response capabilities and timeliness. By combining preset time thresholds to classify lateral risk levels, the system can complete external interactive warnings before critical time points, helping to provide more timely alerts to pedestrians or lateral vehicles, thereby significantly improving the vehicle's active safety protection capabilities and external collaborative interaction level in complex traffic scenarios.

[0015] In some embodiments of this application, the step of identifying rear risks based on acquired rear target motion data to obtain rear risk levels includes: Obtain the relative speed and relative distance of a moving target behind the vehicle relative to the vehicle itself; Determine whether the relative velocity is greater than zero; If the relative speed is not greater than zero, it is determined that there is no risk of collision and the rear risk level is determined. If the relative velocity is greater than zero, the estimated time of rear collision is calculated based on the relative velocity and the relative distance. The rear risk level is determined based on the relationship between the estimated rear collision time and a preset second time threshold.

[0016] In this technical solution, by acquiring the relative speed and relative distance of a moving target behind the vehicle, the system can determine the approaching situation of vehicles approaching from behind in real time. When the relative speed is not greater than zero, meaning the vehicle behind is not approaching, it can be directly determined that there is no risk of collision, thus reducing the computational burden on the system. When the relative speed is greater than zero, the estimated collision time is calculated by combining the relative speed and relative distance, and compared with a preset second time threshold to quantify the risk level. This approach not only considers the combined effect of speed and distance but also introduces a time dimension as a criterion, enhancing the dynamism and accuracy of risk identification.

[0017] In some embodiments of this application, determining the comprehensive risk level based on a preset mapping relationship according to the lateral risk level and the rear risk level includes: Based on the combination of the lateral risk level and the rear risk level, query the preset risk mapping relationship; Based on the correspondence rules between the combination and the comprehensive risk level defined by the risk mapping relationship, a comprehensive risk level that matches the combination of the current lateral risk level and the rear risk level is determined.

[0018] In this technical solution, by combining lateral and rear risk levels and conducting a comprehensive assessment based on a pre-defined risk mapping relationship, collaborative decision-making using multi-source risk information can be achieved. This mapping relationship can cover a variety of possible risk combinations, ensuring the system responds more appropriately to complex traffic interactions. Compared to judging based on a single risk indicator, this solution improves the accuracy and stability of risk assessment through level combination mapping, effectively avoiding over- or under-response to local risks. The output of the comprehensive risk level serves as the core basis for configuring external interaction strategies, enabling vehicles to employ graded and categorized interaction methods under different risk scenarios. This effectively enhances the relevance and adaptability of external information transmission, strengthening the overall collaborative safety capabilities of the traffic environment.

[0019] In some embodiments of this application, the step of controlling and configuring at least one of the external voice module, external lighting module, and external interactive screen according to the comprehensive risk level, and executing the corresponding external interactive control strategy, includes: In response to the overall risk level, a control command corresponding to the overall risk level is generated; Based on the control command, at least one of the vehicle exterior voice module, vehicle exterior lighting module and vehicle exterior interactive screen is driven and configured. The driver configuration includes at least one of the following: Adjust the broadcast content and broadcast mode of the external voice module; adjust the flashing frequency and display pattern of the external light module; adjust the display content and display mode of the external interactive screen.

[0020] The technical solution generates corresponding control commands based on a comprehensive risk level and configures the external voice module, lighting module, and interactive screen to enable the vehicle to proactively and safely interact with external traffic participants. The control strategy flexibly adjusts the broadcast content, light frequency, or interactive interface according to different risk levels, thereby achieving tiered responses and differentiated information delivery. This not only enhances the dynamic adaptability of external interactions but also significantly strengthens the vehicle's information expression capabilities and environmental coordination in yielding and complex traffic scenarios, providing crucial support for achieving safer and smarter traffic interactions.

[0021] In some embodiments of this application, the training process of the multimodal intent prediction model includes: Obtain a training sample set, each training sample including multi-source perception information collected by the sample vehicle ahead, and pedestrian cross-road intention labels corresponding to the multi-source perception information ahead; The aforementioned multi-source perception information is input into the multimodal intent prediction model to be trained; Based on the prediction results of the multimodal intent prediction model on the multi-source perception information ahead and the pedestrian's intention to cross the road label, the model prediction loss value is calculated; The parameters of the multimodal intent prediction model are adjusted based on the model's predicted loss value to obtain the trained multimodal intent prediction model.

[0022] In this technical solution, a training sample set containing multi-source perception information and its corresponding intent labels is constructed. This set is then used to supervise the training of a multimodal intent prediction model, enabling the model to effectively learn the correlation between pedestrian crossing intentions and perception features. By inputting actual collected perception information and calculating the loss value between the model's prediction results and the true labels, the model parameters are adjusted in reverse to continuously optimize the prediction capability. The trained model can be embedded into the vehicle control system to determine in real time whether a pedestrian ahead intends to cross, providing a highly reliable decision-making basis for subsequent risk identification and external interaction control, significantly improving the system's adaptability and safety in real-world road scenarios.

[0023] In a second aspect, the present invention provides a vehicle including a controller configured to implement the vehicle safety control method as described in the first aspect.

[0024] In a third aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the vehicle safety control method as described in the first aspect.

[0025] As can be seen from the above technical solutions, additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of the vehicle safety control method according to the embodiments of this application; Figure 2 This is a schematic diagram of the first process for determining the lateral risk level according to the embodiments of this application; Figure 3 This is a schematic diagram of the second process for determining the lateral risk level according to the embodiments of this application; Figure 4 This is a flowchart illustrating the determination of the subsequent risk level according to the implementation method of this application; Figure 5 This is a schematic diagram of the configuration of the vehicle external voice module according to an embodiment of this application; Figure 6 This is a schematic diagram of the first structure of a vehicle safety control system according to an embodiment of this application; Figure 7 This is a second structural schematic diagram according to an embodiment of this application; Figure 8 This is a schematic diagram of a computer device according to an embodiment of this application.

[0027] In the above figures: 100. Environmental perception module; 200. Central control module; 300. External interaction module; 40. Bus; 41. Processor; 42. Memory; 43. Communication interface. Detailed Implementation

[0028] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between components; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature. In this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0029] The present application will now be described in detail through exemplary embodiments. However, it should be understood that, without further description, elements, structures, and features in one embodiment may be advantageously incorporated into other embodiments. It should be noted that in the field of automotive safety technology, with the rapid development of electronic technology, sensor technology, and artificial intelligence algorithms, the active safety capabilities of vehicles have evolved from non-existent to comprehensive. Early safety technologies were mainly passive protection, such as seat belts and airbags, aimed at mitigating injuries after an accident. With technological advancements, automotive safety has entered a stage of active safety focused on accident prevention.

[0030] Advanced driver assistance systems (ADAS) have become the mainstream direction of current technological development. They acquire real-time environmental information around the vehicle by integrating multiple sensors in the forward, side, and rear views, and use on-board computing units to process, identify, and make decisions, thereby enabling early warning or automatic intervention for potential collision risks.

[0031] In existing technologies, most solutions are still limited to assessing the risk of direct collision between the vehicle and pedestrians, lacking the ability to identify and prevent potential side and rear collisions caused by the vehicle's yielding behavior. Most existing systems primarily target the driver inside the vehicle for warnings, neglecting the necessity of transmitting information from outside the vehicle, resulting in pedestrians and surrounding vehicles being unable to perceive the vehicle's intentions in a timely manner, creating interactive blind spots.

[0032] At the same time, the existing interaction methods are relatively simple and it is difficult to output differentiated information according to different risk levels. This limits the system's adaptability and response accuracy in complex traffic scenarios and makes it difficult to effectively prevent typical problems such as secondary accidents caused by yielding.

[0033] Based on this, this application proposes a vehicle safety control method, system, and electronic device. It acquires multi-source perception information from the front to determine the pedestrian's intention to cross the road. When an intention is determined, it performs parallel lateral and rear risk identification to obtain lateral and rear risk levels, respectively. Then, based on the dual-side risk levels and a preset mapping relationship, it determines a comprehensive risk level. Finally, based on the comprehensive risk level, it controls and configures at least one of the external voice module, external lighting module, and external interactive screen to execute a coordinated external interactive control strategy. This achieves integrated perception, comprehensive decision-making, and multi-directional proactive interaction of the global risks of the complex system consisting of the vehicle, pedestrians, and vehicles behind and to the sides. It solves the problem that existing technologies, which only focus on the single collision relationship between the vehicle and the pedestrian and whose warning measures are only directed towards the inside of the vehicle, cannot effectively address the chain-like and systemic safety risks triggered when yielding to pedestrians, such as rear-end collisions and side blind spot collisions.

[0034] It should be noted that, in order to facilitate understanding of the vehicle safety control system and its external interaction mechanism involved in this invention, the following related terms are explained below.

[0035] In the following, embodiments of this application will be described in detail with reference to the accompanying drawings.

[0036] As attached Figures 1 to 5 As shown, in one illustrative embodiment of the vehicle safety control method, system and electronic device of this application, the vehicle safety control method includes the following steps S1-S4.

[0037] S1: Obtain multi-source perception information ahead, and based on the multi-source perception information ahead, determine whether the target pedestrian has the intention to cross the road through a multi-modal intent prediction model.

[0038] Preferably, the vehicle uses sensors and data processing modules integrated into the vehicle body to collect and integrate various types of data that constitute multi-source perception information in front in real time, specifically including the following information.

[0039] The system acquires location and scene information by using a Global Positioning System (GPS) module and high-precision map data to obtain the vehicle's precise location and heading, and determines whether it is in or near pedestrian crossings, intersections, or other similar areas. It also acquires static environmental semantics such as road speed limits and lane markings.

[0040] Visual information is acquired by capturing continuous image frames from a forward-facing camera, from which road markings such as pedestrian crossings and traffic signs can be extracted, as well as the appearance, outline, and preliminary location of pedestrians.

[0041] Acquire millimeter-wave radar point cloud information, obtain point cloud data of target objects through forward millimeter-wave radar or lidar, and provide accurate distance, radial velocity and orientation information of traffic participants, including pedestrians, especially providing stable detection under visually limited conditions.

[0042] Obtain vehicle status information, including the vehicle's real-time speed, acceleration, steering angle, and other dynamic states.

[0043] By acquiring microscopic temporal information of pedestrians and continuously tracking pedestrian targets detected by vision and millimeter-wave radar, their motion trajectory and speed change sequence can be generated, and computer vision algorithms can be used to estimate their temporal behavioral characteristics such as body orientation, line of sight, and posture.

[0044] After preprocessing such as time synchronization, coordinate system unification, and calibration, the aforementioned multi-source raw data enters a feature extraction and fusion layer. Here, different types of data are converted into a unified feature representation that the model can process.

[0045] For example, images are converted into feature vectors, millimeter-wave radar point clouds are converted into voxel features or target list features, temporal trajectories are encoded into feature sequences, and map semantics and vehicle status are embedded as contextual features. These heterogeneous features are collectively constructed into a structured, semantically rich dynamic representation of the traffic scene ahead.

[0046] Furthermore, this fused scene representation is input into a pre-trained multimodal intent prediction model. The multimodal intent prediction model is built using a deep neural network architecture and has the ability to process image features, radar features, behavioral temporal features, and contextual semantic information.

[0047] Furthermore, the multimodal intent prediction model can be a multimodal temporal transformation (MTT) model based on the Transformer architecture. This model has the ability to process image features, radar features, behavioral temporal features and contextual semantic information simultaneously, and is suitable for reasoning and judging complex behavioral intents in dynamic traffic scenarios.

[0048] The multimodal intent prediction model structure includes the following modules.

[0049] First, in the input encoding layer, image frames captured by the forward-looking camera are processed by pre-trained convolutional neural networks such as ResNet-50 to extract semantic features; point cloud data acquired by millimeter-wave radar is transformed into structured spatial features through voxel networks such as PointPillars; behavioral features such as pedestrian temporal trajectories and posture sequences are encoded using LSTM; at the same time, map semantic information such as road speed limits and lane lines, as well as the vehicle's own state, are embedded as contextual inputs.

[0050] Subsequently, in the multimodal fusion and temporal modeling layer, all modal features are fed into the time-aware Transformer model. The model uses a multi-head self-attention mechanism to model the dependencies between modalities and different time steps, and automatically extracts high-order semantic features and behavioral patterns across modalities.

[0051] Finally, in the classification and determination layer, the fused features are processed by a fully connected layer and a Softmax function to output the probability result of whether the target pedestrian intends to cross the road, which is represented as P = [P0,P1], where P1 represents the probability of "intention to cross the road".

[0052] In one specific embodiment, when the vehicle approaches an urban intersection, the forward-facing camera detects a zebra crossing pattern, and the forward-facing millimeter-wave radar detects a stationary pedestrian approximately 12 meters ahead. The pedestrian remains stationary for more than 2 seconds, facing the road and looking at the vehicle, accompanied by a slight forward lean. The system inputs multimodal data, including images, radar data, behavioral timing, and contextual semantics, into a multimodal intent prediction model for inference. The model outputs P0 = 0.27 and P1 = 0.73, determining that the target pedestrian has a clear intention to cross the road.

[0053] In some embodiments, the multimodal intent prediction model training process includes: Obtain a training sample set, each training sample including the multi-source perception information collected by the sample vehicle ahead, and the pedestrian cross-road intention label corresponding to the multi-source perception information ahead; Input the multi-source perception information from the front into the multimodal intent prediction model to be trained; The prediction loss value of the model is calculated based on the prediction results of the multi-source perception information ahead and the pedestrian's intention to cross the road label. The parameters of the multimodal intent prediction model are adjusted based on the model's predicted loss value to obtain the trained multimodal intent prediction model.

[0054] Preferably, the training process of the multimodal intent prediction model can employ an end-to-end supervised learning framework, combining data collected from real road scenarios with data generated from simulation environments to enhance the model's generalization ability and robustness. The specific implementation steps of this training process are as follows.

[0055] During the data acquisition phase, a training sample set containing multimodal information is obtained. Each sample consists of multi-source perception information collected synchronously by sensors such as forward-facing cameras, millimeter-wave radar, laser millimeter-wave radar, GPS, and high-precision maps during actual vehicle operation.

[0056] Simultaneously, through manual annotation or an automatic behavior tag generation system, each sample is labeled with a classification tag indicating whether the pedestrian has the "intent to cross the road". The samples should cover a variety of typical scenarios, including different weather conditions, lighting, road types, traffic density, and pedestrian status.

[0057] During the model training phase, multi-source perception information from the front undergoes unified preprocessing, feature extraction, and encoding before being input into the multimodal intent prediction model to be trained. The model outputs a prediction result P = [P0, P1], representing the probability distribution of whether the target pedestrian in the sample has or does not have the intent to cross.

[0058] Subsequently, the predicted loss value is calculated using the cross-entropy loss function by analyzing the difference between the model output and the true label of the sample.

[0059] The loss function can be expressed as:

[0060] in, P1 represents the true label of the sample, and P1 is the predicted probability of the multimodal intent prediction model for "there is an intent to cross the road".

[0061] Multimodal intent prediction models use backpropagation and optimizers such as stochastic gradient descent (SGD) or Adam to automatically adjust the parameters of each neural network layer, gradually reducing prediction errors.

[0062] During training, mechanisms such as Dropout and Batch Normalization can be introduced to prevent overfitting, and the generalization performance of the model can be evaluated through the validation set. Indicators such as accuracy, recall, and F1 score can be used to monitor the training effect of the model.

[0063] After training, a deep neural network model with strong temporal modeling and multimodal fusion capabilities is obtained. This model can be deployed in vehicle control systems to perform real-time, high-precision recognition of pedestrians' crossing intentions in real-world road scenarios. The training process supports regular updates and online incremental training to continuously improve the model's adaptability to new scenarios.

[0064] S2: When it is determined that the target pedestrian intends to cross the road, lateral risk identification is performed based on the acquired lateral target movement data and pedestrian movement data to obtain the lateral risk level; and rear risk identification is performed based on the acquired rear target movement data to obtain the rear risk level.

[0065] When it is determined that the target pedestrian intends to cross the road, in order to ensure the pedestrian's safety and avoid potential traffic conflicts or secondary accidents caused by the vehicle's yielding behavior, it is necessary to consider the dynamic interaction between the vehicle and other traffic participants on the side and behind during the stopping or deceleration process.

[0066] Specifically, this includes the interaction risk between the vehicle and vehicles approaching from the side, as well as the following risk between the vehicle and vehicles approaching from behind. Therefore, it is necessary to identify side risks and rear risks separately.

[0067] In some embodiments, such as Figure 2 As shown, the process of identifying lateral risks based on the acquired lateral target motion data and pedestrian motion data, and obtaining the lateral risk level, includes the following steps S21-S24.

[0068] S21: Obtain the first motion state information of the moving target to the side and the second motion state information of the pedestrian.

[0069] Preferably, the initial motion state information of the lateral moving target is acquired by a millimeter-wave radar and / or a side-view camera arranged on the side of the vehicle. The lateral side of the vehicle includes the direction to the left of the vehicle and the direction to the right of the vehicle.

[0070] The side millimeter-wave radar provides precise distance, radial velocity, and azimuth of moving targets to the side relative to the vehicle.

[0071] Side-view cameras provide the target's visual outline, category, and richer motion context information.

[0072] By using multi-sensor fusion processing, the position coordinates, velocity vector, and heading angle information of the moving target to the side are obtained.

[0073] It should be noted that lateral moving targets can be either motor vehicles or non-motor vehicles.

[0074] Furthermore, the pedestrian's second motion state information is obtained through the vehicle's forward vision system, including a forward-facing camera and a forward-facing millimeter-wave radar. The forward-facing camera analyzes image sequences to extract the pedestrian's image position, posture features, and displacement changes, thereby estimating the pedestrian's velocity direction and motion trend in the image plane.

[0075] Forward millimeter-wave radar provides the relative distance, radial velocity, and trend of change between pedestrians and the vehicle, and is particularly suitable for pedestrian detection under visually unstable conditions such as obstruction, low light, or rain and snow.

[0076] By fusing spatiotemporal information from vision and millimeter-wave radar, and combining motion estimation methods such as Kalman filtering, the real-time position coordinates and velocity vectors of pedestrians can be obtained.

[0077] S22: Based on the first motion state information and the second motion state information, predict the first motion trajectory of the lateral moving target and the second motion trajectory of the pedestrian, respectively.

[0078] Preferably, the trajectory prediction is based on the motion state at the current moment and assumes that the target is moving in a straight line at a constant speed within a short prediction window.

[0079] Specifically: The pedestrian's second trajectory Based on its current location and velocity vector Prediction, that is:

[0080] In the above formula, the pedestrian is assumed to move according to the current velocity vector. Maintain uniform linear motion. Within a short prediction window, the expected trajectory is used to simulate the pedestrian's intention to cross the road.

[0081] Using a uniform linear motion model is an effective and reasonable simplification of pedestrian short-term motion behavior, while ensuring the real-time computing performance of the on-board system.

[0082] The first trajectory of a moving target from the side Based on its current location and velocity vector and its heading angle Prediction, that is:

[0083] In the above equation, the lateral moving target is assumed to maintain its currently observed velocity vector. Perform uniform linear motion. Within a short prediction window, it is used to simulate the future trajectory of a moving target from the side in its own lane or current direction of travel. S23: Determine whether there is a conflict point between the first motion trajectory and the second motion trajectory.

[0084] Preferably, two predicted trajectories are calculated. and The intersection point. The following details the situation where the two predicted trajectories have conflict points.

[0085] If there is a unique intersection point, then that intersection point is determined as the conflict point (CP).

[0086] Specifically, this can be accomplished through geometric calculations, analyzing the two predicted lines. and The spatial relationships within a two-dimensional plane are used to determine the benchmark point used for collision time assessment, i.e., the point of impact. The specific calculation process is as follows: Specifically, the two predicted lines are analyzed through geometric calculations. and The spatial relationships within a two-dimensional plane are used to determine the benchmark point used for collision time assessment, i.e., the point of impact. Its coordinates are represented as The specific calculation process is as follows: First, check if the direction vectors of the two lines are parallel. If the direction vectors are parallel, then the two lines are either parallel or coincident, and there is no unique intersection point under the prediction model.

[0087] When two lines are not parallel, solve the vector equation formed by the parametric equations of the two lines:

[0088] in , These are the starting position vectors for the pedestrian and the lateral target, respectively.

[0089] The above vector equations are decomposed into a system of two-dimensional scalar equations and solved to obtain the coordinates of the intersection points. The point of conflict .

[0090] S24: If a conflict point exists, calculate the lateral estimated collision time between the moving target and the pedestrian based on the conflict point, and determine the lateral risk level based on the lateral estimated collision time.

[0091] In some embodiments, calculating the estimated lateral collision time between a laterally moving target and a pedestrian based on the point of collision includes: Based on the conflict point, calculate the first time when the pedestrian arrives at the conflict point; Calculate the second time when a lateral moving target arrives at the point of conflict; The smaller value between the first time and the second time is determined as the expected time of lateral collision; The lateral risk level is determined based on the relationship between the expected lateral collision time and the preset first time threshold.

[0092] Preferably, the lateral estimated collision time is calculated. The specific process is as follows.

[0093] Calculate the first time required for a pedestrian to reach the conflict point .

[0094] Because pedestrians may not be directly facing the point of conflict. The motion therefore requires calculating its velocity in the direction of... Effective components in the direction. The calculation formula is:

[0095] in, Indicates the pedestrian's current location To the point of conflict Euclidean distance, For the speed of the pedestrian, pedestrian velocity vector With pointing vector The angle between them.

[0096] The physical meaning of the above calculation formula is that when a pedestrian reaches the conflict point... The time required is equal to its arrival The straight-line distance is divided by the magnitude of the projection of its velocity along that straight line. In practical numerical calculations, the vector dot product form is usually used for efficient solution.

[0097] Furthermore, the second time required for a lateral moving target to reach the point of conflict is calculated. Since a target moving laterally may not be moving directly towards the point of conflict (CP), it is necessary to calculate the effective component of its velocity in the direction pointing towards CP. The calculation formula is as follows:

[0098] in, This represents the current position of the target to the side. For him to distance, Its speed, Its velocity vector With vector The angle between them.

[0099] The corresponding vector dot product calculation form is:

[0100] Furthermore, the predicted lateral collision time is determined, and the smaller of the two times mentioned above is taken as the mean. ,Right now:

[0101] The above formula represents the time when the pedestrian or the target on the side first arrives at the potential point of conflict, which is the expected lateral collision time.

[0102] Furthermore, based on the calculated lateral estimated collision time The relationship between the risk level and the preset first time threshold is used to determine the lateral risk level.

[0103] Specifically, The value is compared with a set of predefined first-time thresholds to map to discrete risk levels.

[0104] Optionally, the first time threshold includes two: a first emergency threshold and a first warning threshold. These can be set to 3 seconds and 8 seconds, respectively.

[0105] Furthermore, the logic for determining the lateral risk level is as follows: like If so, the lateral risk level is determined to be Emergency. like If so, the lateral risk level is determined to be a warning. like If so, the lateral risk level is determined to be safe.

[0106] It should be noted that the aforementioned first-time threshold is merely an exemplary value and is not intended to limit the scope of protection of this application. In actual system configurations, the threshold can be dynamically adjusted or calibrated based on vehicle performance, road adhesion coefficient, weather conditions, legal and regulatory requirements, or different driving modes. For example, on slippery roads, the warning threshold may be set earlier.

[0107] In some embodiments, such as Figure 3As shown, determining whether there is a conflict point between the first motion trajectory and the second motion trajectory includes the following steps S25-S26.

[0108] S25: If there is no conflict point, calculate the shortest distance between the first motion trajectory and the second motion trajectory within the prediction time window.

[0109] Preferably, when two predicted trajectories are determined... and There are no intersections, i.e., no points of conflict. At that time, the two will be calculated within a preset prediction time window. Shortest distance within And the corresponding nearest neighbor pairs.

[0110] The specific calculation process is as follows: Establish parameterized trajectory equations: Represent the uniform linear motion trajectories of the pedestrian and the side target in scalar parameter form: Pedestrian trajectory: ,in For time parameters; Trajectory of lateral target: ,in This is a time parameter.

[0111] Construct a distance function and combine it at any time. The squared distance function between two points is as follows:

[0112] Seeking minimum point This will give you the points that are closest to each other and the shortest distance between the two trajectories. .

[0113] Finding the shortest distance time: by setting the squared distance function For parameters and The partial derivatives are zero:

[0114] The following system of linear equations is obtained:

[0115] Solving this system of equations yields the parametric solution that minimizes the distance. and That is, the moment when corresponding points on the two trajectories are closest to each other.

[0116] Calculate the nearest point and the shortest distance: Will Substitution To obtain the nearest point on the pedestrian's trajectory .

[0117] Will Substitution Obtain the nearest point on the trajectory of the target from the side. .

[0118] Calculate the Euclidean distance between two points; this is the shortest distance. .

[0119] S26: Determine the lateral risk level based on the relationship between the shortest distance and the preset safe distance threshold.

[0120] Furthermore, based on the calculated shortest distance The relationship with the preset safe distance threshold, which can be set to 3 meters, determines the lateral risk level in this situation: like The risk level is determined to be no collision risk and the side risk level is Safe. like If the lateral risk level is determined to be high-risk and close-range, even if the predicted paths do not intersect, it may cause danger or pedestrian panic. Therefore, the lateral risk level should be classified as a warning.

[0121] The above calculation process provides an accurate spatial risk measurement for scenarios where trajectories do not intersect, ensuring the completeness and robustness of the lateral risk identification module.

[0122] It should be noted that the preset safe distance threshold is only an exemplary value for a typical embodiment. In actual system implementation, this threshold can be dynamically configured or adaptively adjusted based on factors such as vehicle type, road grade, weather conditions, pedestrian behavior model, traffic regulations, and the user's acceptable level of comfort.

[0123] For example, in rainy or foggy weather with low visibility or in school areas, the threshold can be appropriately increased to provide a more conservative safety guarantee; while in environments with wide roads and good visibility, the standard threshold can be used.

[0124] In some embodiments, such as Figure 4 As shown, the process of identifying rear risks based on the acquired rear target motion data and obtaining the rear risk level includes the following steps S27-S29.

[0125] S27: Obtain the relative speed and relative distance of a moving target behind the vehicle relative to the vehicle; determine whether the relative speed is greater than zero.

[0126] Preferably, the relative speed of the moving target behind. and relative distance The data is directly acquired by a millimeter-wave radar positioned at the rear of the vehicle. As the core sensor, the rear-facing millimeter-wave radar can stably and reliably measure the radial relative velocity and absolute distance of targets behind the vehicle. Its performance is less affected by weather and lighting conditions, providing an accurate data foundation for rear risk identification.

[0127] Furthermore, the real-time relative speed (Vrear) determines whether the relative speed is greater than zero only if the approaching speed of the target behind is greater than the vehicle's speed. Only then is a rear-end collision theoretically possible. If... This indicates that the vehicle behind is traveling at the same speed or lower, and there is no risk of approaching.

[0128] S28: If the relative speed is not greater than zero, it is determined that there is no risk of collision and the rear risk level is determined; if the relative speed is greater than zero, the estimated rear collision time is calculated based on the relative speed and relative distance.

[0129] Preferably, when At that time, it was determined that there was no risk of collision, and the estimated time for a collision from behind was [not specified]. Set a predetermined constant, such as 500 seconds, that is much larger than the actual possible collision time. This constant value is much larger than any risk time threshold defined by the system, thus ensuring that the situation will be consistently classified as the lowest risk level in subsequent comparisons.

[0130] Preferably, when determining At that time, the estimated time of the rear collision Calculate directly using the following formula:

[0131] In the above formula, the current relative velocity is used. To get closer, one must overcome the current relative distance. The required time, i.e. the estimated time of the rear-end collision.

[0132] S29: Determine the rear risk level based on the relationship between the estimated rear collision time and the preset second time threshold.

[0133] Preferably, the calculated The risk level is determined by comparing the data with a set of preset second time thresholds.

[0134] Optionally, the second time threshold includes two: a second emergency threshold and a second warning threshold. These can be set to 3 seconds and 8 seconds, respectively.

[0135] Furthermore, the risk level in the rear is determined as follows: like If so, the risk level behind is determined to be Emergency. like If so, the risk level behind is determined to be a warning. like If so, the risk level behind is determined to be safe.

[0136] It should be noted that the aforementioned second time threshold is merely an exemplary value and is not intended to limit the scope of protection of this application. In actual system configurations, the threshold can be dynamically adjusted or calibrated based on vehicle performance, road adhesion coefficient, weather conditions, legal and regulatory requirements, or different driving modes. For example, on slippery roads, the second warning threshold may be set earlier.

[0137] S3: Determine the overall risk level based on the lateral risk level and the rear risk level, using a preset mapping relationship.

[0138] In some embodiments, determining the overall risk level based on a preset mapping relationship, according to the lateral risk level and the rear risk level, includes: Based on the combination of lateral and rear risk levels, query the preset risk mapping relationship; Based on the correspondence rules between combinations and comprehensive risk levels defined by the risk mapping relationship, determine the comprehensive risk level that matches the combination of the current lateral risk level and the rear risk level.

[0139] Preferably, the overall risk level can be determined by querying a predefined risk decision matrix, as shown in Table 1.

[0140] The risk decision matrix establishes a deterministic mapping between all possible combinations of lateral and rear risk levels and a comprehensive risk level, and associates specific collaborative response strategies.

[0141] The lateral and posterior risk levels are normalized according to the level standards defined in the matrix and used as query keys. Based on the current combination of lateral and posterior risk levels, the risk decision matrix is ​​queried, and the comprehensive risk level that uniquely corresponds to this combination in the matrix is ​​the final risk assessment result.

[0142]

[0143] Table 1 Risk Decision Matrix S4: Based on the comprehensive risk level, control configuration is performed on at least one of the vehicle external voice module, vehicle external lighting module, and vehicle external interactive screen, and the corresponding vehicle external interactive control strategy is executed.

[0144] In some embodiments, based on the comprehensive risk level, control configuration is applied to at least one of the external voice module, external lighting module, and external interactive screen, and the corresponding external interactive control strategy is executed, including: In response to the overall risk level, generate control instructions corresponding to the overall risk level; Based on control commands, at least one of the vehicle external voice module, vehicle external lighting module and vehicle external interactive screen is driven and configured. The driver configuration includes at least one of the following: Adjust the broadcast content and mode of the external voice module; adjust the flashing frequency and display pattern of the external light module; adjust the display content and display mode of the external interactive screen.

[0145] Preferably, the specific implementation process of executing the external interaction control strategy based on the comprehensive risk level is as follows: Based on the comprehensive risk level determined from the risk decision matrix and its associated system response decisions, a specific and executable set of control instructions is generated.

[0146] The control command set is distributed to the corresponding external interaction module controller through the vehicle's internal network, driving it to execute coordinated warnings.

[0147] Furthermore, such as Figure 5 As shown, the specific steps for configuring the driver for the vehicle's external voice module are as follows.

[0148] Based on the final determined comprehensive risk level, the central control unit generates a digital control message containing the target risk scenario and corresponding voice commands. This control message is broadcast through the vehicle's internal Controller Area Network (CAN) bus.

[0149] The Head Unit (HUT), located inside the vehicle's cabin, serves as a key hub for interaction with the outside world. Its internal Microcontroller Unit (MCU) continuously monitors the CAN bus.

[0150] Once the MCU recognizes and captures the risk control message sent to it, it immediately parses it and extracts the key control parameters. The MCU then routes the parsed control data at high speed and reliably to the dedicated audio processing subsystem in the HUT's internal System-on-Chip (SoC) via its Serial Peripheral Interface (SPI) bus with the main processor.

[0151] Upon receiving a command, the SOC audio module activates its built-in external voice alert function. This function indexes and retrieves the corresponding voice segment from a pre-stored high-quality audio library based on the input risk level code. For example, it might retrieve the "Danger! Stop!" audio file corresponding to "Extremely Dangerous," or generate a corresponding prompt statement in real-time using a text-to-speech (TTS) engine.

[0152] The generated raw digital audio stream is then fed into a digital signal processor (DSP) integrated into or tightly coupled to the SOC. The DSP performs optimization processing on the audio stream according to preset audio processing algorithms, including decoding, sampling rate conversion, equalization adjustment, and dynamic volume control, to ensure clear audio quality and that the warning intensity matches the risk level.

[0153] The final audio data, processed by the DSP, is output through a Time Division Multiplexing (TDM) digital audio interface. The TDM interface allows multiple audio channels to be transmitted on the same set of data lines; in this embodiment, it is used to independently drive Class D audio power amplifier channels connected to different physical speaker units. Each amplifier channel efficiently drives exterior speakers with a power output of at least 15W, located within the vehicle's front bumper or behind the grille.

[0154] Ultimately, the speaker converts the electrical signal into sound waves, broadcasting a clear and powerful directional warning or prompt to the outside environment, thus completing the entire process from risk identification to voice-interactive alarm.

[0155] Furthermore, the specific steps for configuring the driver for the vehicle's external interactive screen are as follows.

[0156] The central control unit generates control commands that include display mode, flashing frequency, and graphic content based on the overall risk level and specific risk direction.

[0157] The command is sent to the vehicle's Head Unit (HUT) via a high-speed in-vehicle network. The graphics processing unit within the HUT parses and renders the command, generating corresponding video frames or graphic commands, and then sends them to the rear-view interactive screen controller located at the rear of the vehicle via the Ethernet video protocol.

[0158] The rear interactive screen adopts an intelligent pixelated LED matrix design, with a physical resolution of 32 (height) × 128 (width) and a pixel pitch of 3mm. It is controlled by a row and column scanning driver IC and supports PWM dimming to achieve rich dynamic graphics display and precise brightness control.

[0159] Optionally, the controller drives the screen according to the received instructions.

[0160] When the overall risk level is "highest" or "high" and the risk originates directly behind, the controller drives the entire screen to flash at a high frequency of 3Hz, while simultaneously overlaying a real-time updated "distance bar" graphic on its display area. The length or fill ratio of this light bar is inversely proportional to the real-time calculated estimated rear collision time, intuitively conveying the urgency of the approach to following vehicles.

[0161] When the risk primarily originates from the side or rear, the controller maintains necessary warning flashing while simultaneously displaying a dynamic red "arrow pulse" pattern on the left or right edge of the screen. The arrow points in the same direction as the source of the risk, clearly informing vehicles to the side or rear to be aware of pedestrians in their blind spots.

[0162] When the risk level is "medium" or "low", the controller drives the screen to flash at a low frequency or display static, mild alert icons until the risk is eliminated.

[0163] Furthermore, the specific steps for configuring the driver for the vehicle's exterior lighting module are as follows.

[0164] The drive commands for the exterior lighting modules are typically sent from the HUT or central controller to the body control module (BCM) or dedicated lighting controller via the CAN bus. These commands contain specific flashing mode codes tailored to the current risk level.

[0165] When the system determines that there is a high level of risk such as "extreme danger", "emergency behind", or "emergency to the side", the controller drives the left and right turn signals to flash synchronously at a high frequency of 3Hz, generating a strong and eye-catching visual warning signal.

[0166] When the risk level is determined to be medium, such as "rear warning", the hazard lights will be activated to "flash rapidly" at a relatively low frequency of 1.5Hz to provide an effective warning while distinguishing it from higher levels of urgency.

[0167] Furthermore, the driving of the lighting module and the display on the rear interactive screen work in tandem. For example, in a "side emergency" scenario, in addition to the hazard lights flashing, the system also provides directional guidance through arrow pulses on the interactive screen; in a "rear emergency" scenario, the hazard lights flashing and the distance bar display on the interactive screen are combined to form a composite warning with superimposed intensity and information.

[0168] The above-described driving configurations for the vehicle's external interactive screen and external lighting modules, combined with the driving configuration for the external voice module, enable the system to output multi-dimensional and complementary warning information to pedestrians, vehicles behind, and vehicles to the side and rear by synchronously and differentiatedly controlling different external devices. This creates a three-dimensional active safety interactive barrier in complex mixed pedestrian and vehicle traffic scenarios.

[0169] In a second aspect, the present invention provides a vehicle, such as Figures 6-7 As shown, the vehicle safety control method described above includes an environmental perception module 100, a central control module 200, and an external interaction module 300.

[0170] The environmental perception module 100 is used to acquire multi-source perception information in front, side target motion data, pedestrian motion data and rear target motion data; The central control module 200 is communicatively connected to the environmental sensing module 100 for: When it is determined that the target pedestrian intends to cross the road, lateral risk identification is performed based on the acquired lateral target movement data and pedestrian movement data to obtain the lateral risk level; and rear risk identification is performed based on the acquired rear target movement data to obtain the rear risk level. The overall risk level is determined based on the lateral risk level and the rear risk level, according to a preset mapping relationship. The external interaction module 300 is communicatively connected to the central control module 200 and is used to control and configure at least one of the external voice module, external lighting module and external interactive screen according to the comprehensive risk level, and execute the corresponding external interaction control strategy.

[0171] Combination Figure 8 As shown, this embodiment discloses a specific implementation of a computer device. The computer device may include a processor 41 and a memory 42 storing computer program instructions.

[0172] Specifically, the processor 41 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0173] The memory 42 may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory 42 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 42 may include removable or non-removable (or fixed) media. Where appropriate, the memory 42 may be internal or external to the data processing device. In a particular embodiment, the memory 42 is non-volatile. Volatile memory. In a particular embodiment, memory 42 includes read-only memory. ROM (ROM-only memory) and RAM (Random Access Memory). Where appropriate, the ROM can be a mask-programmed ROM or a programmable ROM. Only Memory (PROM) and Erasable Programmable Read-Only Memory (EPRROM) The RAM can be a type of RAM, such as EPROM (Electrically Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), EAROM (Electrically Alterable Read-Only Memory), or FLASH (Flash Memory), or a combination of two or more of these. Where appropriate, the RAM can be a Static Random Access Memory (SRAM). Access Memory (SRAM) or Dynamic Random Access Memory (DRAM) can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0174] The memory 42 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 41.

[0175] The processor 41 implements the vehicle safety control method in the above embodiments by reading and executing computer program instructions stored in the memory 42.

[0176] In some embodiments, the computer device may further include a communication interface 43 and a bus 40. For example, Figure 8 As shown, the processor 41, memory 42, and communication interface 43 are connected through bus 40 and complete communication with each other.

[0177] Communication interface 43 is used to enable communication between modules, devices, units and / or equipment in the embodiments of this application.

[0178] Communication port 43 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0179] Bus 40 includes hardware, software, or both, that couples components of a computer device together. Bus 40 includes, but is not limited to, at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. For example, and not as a limitation, bus 40 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, and a PCI bus. Express (PCI X) bus, Serial Advanced Technology Accessory (Seria l Advanced) The bus may be a Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 40 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0180] Furthermore, in conjunction with the vehicle safety control methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the vehicle safety control methods in the above embodiments.

[0181] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described vehicle safety control method.

[0182] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described vehicle safety control method.

[0183] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A vehicle safety control method characterized by, It comprises: acquiring front multi-source perception information, based on the front multi-source perception information, judging whether the target pedestrian has a crossing road intention through a multi-modal intention prediction model; when it is determined that the target pedestrian has the crossing road intention, performing side risk identification based on the acquired side target motion data and pedestrian motion data to obtain a side risk level; and performing rear risk identification based on the acquired rear target motion data to obtain a rear risk level; determining a comprehensive risk level based on a preset mapping relationship according to the side risk level and the rear risk level; according to the comprehensive risk level, performing control configuration on at least one of an off-vehicle voice module, an off-vehicle light module and an off-vehicle interaction screen, and executing a corresponding off-vehicle interaction control strategy.

2. The vehicle safety control method according to claim 1, characterized by, The side risk identification based on the acquired side target motion data and pedestrian motion data to obtain a side risk level comprises: acquiring first motion state information of a side motion target and second motion state information of the pedestrian; based on the first motion state information and the second motion state information, predicting a first motion trajectory of the side motion target and a second motion trajectory of the pedestrian respectively; judging whether there is a conflict point between the first motion trajectory and the second motion trajectory; if there is a conflict point, calculating a lateral expected collision time between the side motion target and the pedestrian based on the conflict point, and determining the side risk level based on the lateral expected collision time.

3. The vehicle safety control method according to claim 2, characterized by The judgment of whether there is a conflict point between the first motion trajectory and the second motion trajectory comprises: if there is no conflict point, calculating the shortest distance between the first motion trajectory and the second motion trajectory within a prediction time window; determining the side risk level according to the relationship between the shortest distance and a preset safety distance threshold.

4. The vehicle safety control method according to claim 2, characterized by The calculation of the lateral expected collision time between the side motion target and the pedestrian based on the conflict point comprises: based on the conflict point, calculating a first time for the pedestrian to reach the conflict point; calculating a second time for the side motion target to reach the conflict point; determining the smaller value between the first time and the second time as the side expected collision time; determining the side risk level according to the relationship between the side expected collision time and a preset first time threshold.

5. The vehicle safety control method according to claim 1, characterized by The rear risk identification based on the acquired rear target motion data to obtain a rear risk level comprises: acquiring a relative speed and a relative distance of a rear motion target relative to the vehicle; judging whether the relative speed is greater than zero; if the relative speed is not greater than zero, determining that there is no collision risk and determining the rear risk level; if the relative speed is greater than zero, calculating a rear expected collision time based on the relative speed and the relative distance; determining the rear risk level according to the relationship between the rear expected collision time and a preset second time threshold.

6. The vehicle safety control method according to claim 1, characterized by The determination of the comprehensive risk level based on the preset mapping relationship according to the side risk level and the rear risk level comprises: querying a preset risk mapping relationship according to the combination of the side risk level and the rear risk level; According to a corresponding rule of combination and comprehensive risk level defined by the risk mapping relationship, a comprehensive risk level matched with the combination of the current side risk level and the rear risk level is determined.

7. The vehicle safety control method according to claim 1, characterized by According to the comprehensive risk level, at least one of an out-of-vehicle voice module, an out-of-vehicle light module, and an out-of-vehicle interactive screen is configured for control, and a corresponding out-of-vehicle interactive control strategy is executed, including: In response to the comprehensive risk level, a control instruction corresponding to the comprehensive risk level is generated; Based on the control instruction, at least one of the out-of-vehicle voice module, the out-of-vehicle light module, and the out-of-vehicle interactive screen is configured for driving; The driving configuration includes at least one of the following: Adjust the broadcast content and broadcast mode of the out-of-vehicle voice module; adjust the flashing frequency and display pattern of the out-of-vehicle light module; adjust the display content and display mode of the out-of-vehicle interactive screen.

8. The vehicle safety control method according to claim 1, characterized by The multi-modal intention prediction model training process includes: Obtain a training sample set, each training sample including front multi-source perception information collected by a sample vehicle, and a pedestrian crossing road intention annotation label corresponding to the front multi-source perception information; Input the front multi-source perception information into the multi-modal intention prediction model to be trained; Based on the prediction result of the multi-modal intention prediction model on the front multi-source perception information and the pedestrian crossing road intention annotation label, a model prediction loss value is calculated; According to the model prediction loss value, the parameters of the multi-modal intention prediction model are adjusted to obtain a trained multi-modal intention prediction model.

9. A vehicle characterized by comprising: The controller is configured to implement the vehicle safety control method of any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the vehicle safety control method of any one of claims 1 to 8.

Images