Vehicle-based risk warning method, device and equipment and storage medium

Abstract

The application relates to the technical field of intelligent networked vehicle safety assistance, in particular to a risk early warning method and device based on a vehicle, equipment and a storage medium, and aims to solve the problems of existing blind area early warning relying on single detection or fixed threshold, discontinuous risk assessment, insufficient real-time and adaptability to different drivers. The method probabilistically detects the possibility of a blind area event occurring within a preset time period during vehicle driving, and when the probability reaches a preset condition, generates corresponding risk early warning information in combination with personalized data representing the historical driving experience of the driver. By converting the blind area danger from a static judgment to a continuous probability evaluation and introducing a differentiated adaptation mechanism for drivers, the accuracy of blind area risk assessment, the rationality of early warning timing and the pertinence of prompt content can be improved, thereby improving the driving safety assistance effect in complex traffic environments.

Description

Technical Field

[0001] This application relates to the field of intelligent connected vehicle safety assistance, and in particular to a vehicle-based risk warning method, device, equipment and storage medium. Background Technology

[0002] Current vehicle blind spot warning systems mostly use radar, ultrasound, or cameras to collect rearward information and issue prompts to the driver based on preset rules.

[0003] However, such solutions are usually based on static judgments using local perception results, making it difficult to continuously assess risks by combining vehicle operating status, resulting in insufficient accuracy and real-time performance of warnings in complex scenarios; at the same time, the uniform prompt format has limited adaptability to different drivers, affecting the effective reception and response to warning information.

[0004] Therefore, improving the accuracy, real-time performance, and adaptability to different drivers in blind spot warning scenarios has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a vehicle-based risk warning method, apparatus, device, and storage medium to address the aforementioned technical problems. The method targets vehicle blind spot risk warning scenarios. By characterizing the probability of blind spot events occurring within a preset time period and combining this with driver-related characteristics to specifically determine the warning content, it achieves dynamic assessment and differentiated warnings of blind spot risks, improving the effectiveness of risk warnings in complex scenarios.

[0006] In a first aspect, embodiments of this application provide a vehicle-based risk warning method, including:

[0007] The probability of detecting a vehicle's blind spot is defined as the probability of a blind spot event occurring within a preset time period. The blind spot event is defined as an event in which a dynamic target exists within the vehicle's blind spot.

[0008] If the blind spot probability is greater than a preset probability threshold, a target warning message is generated based on the blind spot probability and preset personalized data; the personalized data represents the driver's historical driving experience; the target warning message represents a risk warning prompt for blind spot hazards.

[0009] In one possible implementation, detecting the probability of a vehicle's blind spot includes:

[0010] Acquire blind spot image data of a vehicle within a preset time period; the blind spot image data represents the road conditions within the vehicle's blind spot.

[0011] Based on the blind spot image data, the probability of the vehicle's blind spot is determined.

[0012] In one possible implementation, determining the blind spot probability of a vehicle based on the blind spot image data includes:

[0013] Based on the blind spot image data within each preset time window, the marking information corresponding to each preset time window is determined; the marking information indicates whether a blind spot event has occurred within the preset time window; the preset time window is located within the preset time period.

[0014] The probability of a vehicle's blind spot is determined based on the marker information corresponding to each preset time window.

[0015] In one possible implementation, the method further includes:

[0016] Acquire remote data; the remote data represents the road conditions ahead of the vehicle; the distance between the vehicle and the road ahead is greater than a preset distance threshold;

[0017] Based on the remote data, a first risk status is determined; the first risk status indicates whether there is a driving risk on the road ahead.

[0018] Based on the blind spot probability, a second risk state is determined; the second risk state characterizes the change in the vehicle's blind spot probability relative to the historical blind spot probability.

[0019] If the first risk state indicates a driving risk on the road ahead, and the second risk state indicates that the vehicle's blind spot probability is greater than the historical blind spot probability, then strategy information is generated based on the first risk state and the second risk state; the strategy information represents the vehicle's risk avoidance strategy.

[0020] In one possible implementation, target warning information is generated based on the blind zone probability and preset personalized data, including:

[0021] Based on the personalized data, experience information is determined; the experience information represents the driver's proficiency in driving the vehicle.

[0022] The risk level is determined based on the blind spot probability and the empirical information.

[0023] Based on the risk level, generate target early warning information.

[0024] In one possible implementation, generating target early warning information based on the risk level includes:

[0025] Based on the risk level, an initial warning message is generated; the initial warning message represents a preliminary risk warning message for blind spot hazards.

[0026] Detect the driver's emotional state;

[0027] Based on the emotional state, the initial prompt information is adjusted to obtain the target warning information.

[0028] In one possible implementation, based on the risk level, initial warning information is generated, including:

[0029] The risk level is input into a preset model to obtain the initial prompt information output; the preset model is a trained artificial intelligence model used to generate initial prompt information based on learning from the user's daily conversation content and according to the risk level.

[0030] In one possible implementation, based on the risk level, initial warning information is generated, including:

[0031] Obtain customized information; the customized information represents the perception parameters of the voice interaction preset by the driver.

[0032] Based on the risk level and the customized information, an initial prompt message is generated.

[0033] In one possible implementation, if the blind spot probability is greater than a preset probability threshold, an adjustment command is generated based on the blind spot probability; the adjustment command is used to adjust the driving environment inside the vehicle.

[0034] Secondly, embodiments of this application provide a vehicle-based risk warning device, comprising:

[0035] A detection module is used to detect the probability of a vehicle's blind spot; the blind spot probability represents the probability of a blind spot event occurring within a preset time period; the blind spot event represents an event in which a dynamic target exists within the vehicle's blind spot.

[0036] The generation module is used to generate target warning information based on the blind spot probability and preset personalized data if the blind spot probability is greater than a preset probability threshold; the personalized data represents the driver's historical driving experience; and the target warning information represents a risk warning prompt for blind spot hazards.

[0037] Thirdly, embodiments of this application provide a vehicle-based risk warning device, including: a memory and a processor;

[0038] The memory stores computer-executed instructions;

[0039] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0041] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0042] The vehicle-based risk warning method, device, equipment, and storage medium provided in this application detect the probability of a vehicle's blind spot and, when the probability of the blind spot exceeds a preset probability threshold, generate target warning information based on the blind spot probability and preset personalized data representing the driver's historical driving experience. This enables more targeted warning judgments and prompts for the risk of dynamic targets existing in the vehicle's blind spot, thereby improving the accuracy, real-time performance, and adaptability to different drivers in vehicle blind spot risk assessment, and enhancing the effective reception and response of warning information. Attached Figure Description

[0043] 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.

[0044] Figure 1 A flowchart illustrating the vehicle-based risk warning method provided in this application. Figure 1 ;

[0045] Figure 2 A flowchart illustrating the vehicle-based risk warning method provided in this application. Figure 2 ;

[0046] Figure 3 A schematic diagram of the structure of the vehicle-based risk warning device provided in this application;

[0047] Figure 4 This is a schematic diagram of the structure of the vehicle-based risk warning device provided in this application.

[0048] 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

[0049] 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.

[0050] Vehicle safety assistance and driving risk warning technologies are widely used in passenger cars, commercial vehicles, and road vehicles with intelligent connectivity capabilities. They are particularly suitable for highway driving, lane changes on urban expressways, nighttime driving, rainy and foggy weather, tunnel sections, and complex traffic environments with significant obstruction from large trucks. In these scenarios, drivers need to continuously monitor the area behind and to the sides of the vehicle for rapidly approaching vehicles, motorcycles, non-motorized vehicles, or pedestrians. This area is often beyond the complete coverage of traditional rearview mirrors or the driver's line of sight, thus easily creating blind spot risks.

[0051] To support blind spot warning functions, existing vehicles are typically equipped with rearward sensing hardware, such as rearview mirror cameras, side and rear cameras, millimeter-wave radar, ultrasonic sensors, or body controllers. The onboard processing unit analyzes the collected rearward environmental data and then issues reminders to the driver through the instrument panel, buzzer, central control screen, or indicator lights.

[0052] From a system architecture perspective, this type of solution mainly relies on the vehicle's local perception link, where sensors are responsible for sampling, the onboard controller is responsible for judgment, and the in-vehicle alert module is responsible for output, forming a relatively closed single-vehicle warning loop. Although this technology has already played a basic auxiliary role in daily driving, with the increasing complexity of road traffic environments and the growing individual differences among drivers, relying solely on single local perception results for risk alerts is becoming increasingly difficult to meet the actual needs of continuous assessment, accurate warnings, and adaptive alerts in complex scenarios.

[0053] The basic idea of ​​existing vehicle blind spot warning schemes is usually to use radar, ultrasonic waves or cameras to detect the area behind or to the side of the vehicle in real time. When a target is detected to enter the preset monitoring area, or when the relative distance or relative speed between the target and the vehicle meets the preset threshold, the system determines that there is a potential danger and triggers the corresponding warning signal.

[0054] For example, camera solutions typically identify the presence of vehicle outlines, lights, or pedestrian targets behind the vehicle based on the current image frame or a short sequence of images, while radar solutions calculate the target's position and speed based on the target's echo information, and ultrasonic solutions are more often used for obstacle perception in low-speed, close-range scenarios.

[0055] After acquiring the aforementioned perception results, the controller typically outputs a standardized alarm according to preset rules, such as illuminating the rearview mirror warning lights, emitting a buzzer, or displaying a fixed prompt to remind the driver to be cautious when changing lanes or to pay attention to vehicles approaching from behind. This technical approach has some practicality in simple operating conditions, but it reveals significant limitations in complex driving environments.

[0056] Firstly, existing solutions often rely on whether a target is detected at a certain moment as the basis for judgment, or trigger warnings based solely on simple thresholds. They lack continuous quantification of the trend of blind spot events within a preset time period, making it difficult to accurately characterize the accumulation and variability of risks. Consequently, in situations such as vehicles rapidly approaching from behind or repeated short-term obscuration of the target, false alarms, missed alarms, or delayed warnings are likely to occur.

[0057] Secondly, existing solutions do not adequately consider driver differences, typically employing uniform triggering conditions and warning content without taking into account drivers' past driving experience to differentiate risk handling. For example, less experienced drivers often require earlier and clearer risk warnings when facing the same blind spot situation, while more experienced drivers may be more accustomed to a simpler warning method. A uniform warning not only weakens the specificity of the warnings but may also reduce drivers' trust in the warning information and their response efficiency.

[0058] Third, in real road environments, vehicle operating status and traffic flow changes are highly dynamic. Simply relying on local backward perception results for static judgments often fails to reliably reflect the actual level of danger, especially in scenarios such as high-speed lane changes, continuous following, and rapid approach of following vehicles. If the probability of blind spot events cannot be represented more continuously, it will be difficult to support the generation of more reasonable risk warnings.

[0059] Therefore, while existing technologies can provide basic blind spot warnings, there is still considerable room for improvement in terms of the accuracy of risk assessment, the real-time nature of warnings, and adaptability to different drivers.

[0060] In view of this, how to improve the accuracy, real-time performance, and adaptability to different drivers in vehicle blind spot warning scenarios has become an urgent technical problem to be solved.

[0061] To address the aforementioned issues, a vehicle-based risk warning method is provided. This method is designed for a local blind spot warning architecture. During vehicle operation, the blind spot probability is first detected. The blind spot probability characterizes the probability of a blind spot event occurring within a preset time period, while a blind spot event characterizes the presence of a dynamic target within the vehicle's blind spot. When the detected blind spot probability exceeds a preset probability threshold, target warning information is generated based on this probability and preset personalized data. The personalized data represents the driver's historical driving experience, and the target warning information provides risk warnings for blind spot hazards. In other words, this approach goes beyond simply judging whether a single blind spot target appears; it transforms blind spot risk into measurable probabilistic information and further combines this with the driver's historical driving experience to create a more tailored warning output. Based on this processing logic, a closer correlation can be established between the warning information, the degree of blind spot hazard, and individual driver differences. This improves the targeting and effectiveness of blind spot risk warnings on top of the vehicle's existing perception, processing, and warning chain, laying the technological foundation for more reliable vehicle safety assistance in the future.

[0062] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are 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 now be described with reference to the accompanying drawings.

[0063] Figure 1 A flowchart illustrating the vehicle-based risk warning method provided in this application. Figure 1 ,like Figure 1 As shown, the method includes:

[0064] S101: Detect the blind spot probability of the vehicle; the blind spot probability represents the probability of a blind spot event occurring within a preset time period; the blind spot event represents an event in which a dynamic target exists in the vehicle's blind spot.

[0065] Blind spot probability is used to characterize the probability of a blind spot event occurring within a preset time period. Essentially, it is the result of continuously quantifying the risk level of blind spots to the side or rear of the vehicle, which is different from the instantaneous judgment made based solely on whether a target is detected at a single moment.

[0066] Blind spot events are used to characterize events where dynamic targets exist within the vehicle's blind spot. Dynamic targets can be objects whose position changes relative to the road background, including vehicles approaching from behind, vehicles in adjacent lanes, motorcycles, non-motorized vehicles, and pedestrians. As long as the object shows identifiable displacement, speed changes, or trajectory extensions in continuous sampling images, it can be determined to meet the identification conditions of blind spot events.

[0067] It should be noted that the execution entity in the embodiments of this application can be an on-board processing unit, domain controller, driver assistance controller, or processor that is communicatively connected to the rear perception module, etc., located in the vehicle.

[0068] For example, the executing entity first receives blind spot image data from a rearview mirror camera, a side rear camera, or a rear wide-angle camera, and determines the blind spot detection range by combining the vehicle's current speed, turn signal status, gear information, or vehicle posture information.

[0069] To make blind spot assessment more closely reflect actual driving needs, the system pre-establishes left and right blind spot areas in the vehicle coordinate system. The blind spot areas correspond to the spatial range that is not fully covered by the rearview mirror and is not easily observed by the driver when turning their head. When the vehicle is driving straight, preparing to change lanes, or turning, the processing unit can dynamically adjust the lateral and longitudinal boundaries of the blind spot areas to make the sampling area consistent with the actual risk concern area of ​​the vehicle.

[0070] The probability of detecting a vehicle's blind spot can be achieved by continuously capturing images within the vehicle's blind spot and performing statistical analysis over a preset time period. For example, the system captures a video stream of the blind spot at a fixed frame rate, performs target detection, frame correlation, and motion state recognition on each frame, thereby determining whether a dynamic target exists in the current frame.

[0071] Target detection can be achieved using convolutional neural network models, temporal detection models, or motion detection algorithms based on optical flow and background subtraction. When the center position, bounding box position, or feature vector of the same target in consecutive frames meets the association conditions, the target can be considered to exist continuously in the time dimension. Furthermore, based on the change in position, velocity vector, or trajectory length, it can be determined that it is a dynamic target rather than a static roadside object. If a dynamic target located within the blind zone is identified in a frame or a sequence of consecutive frames, the corresponding moment can be recorded as a blind zone event.

[0072] As an example, the preset time period can be set to a sliding time window of 1 to 10 seconds. The setting of the time window length is used to balance response speed and statistical stability: when the time window is shorter, the system can reflect risk changes more quickly, which is suitable for fast dynamic scenarios such as high-speed lane changes; when the time window is appropriately extended, it can reduce probability jitter caused by short-term occlusion, light flicker, or single-frame false detection, thereby improving the stability of risk assessment.

[0073] To obtain the blind zone probability, the system can perform frame-by-frame statistics on blind zone events within a preset time window. Assuming that blind zone events are detected in M ​​out of a total of N frames, the blind zone probability P can be expressed as P = M / N. Considering the duration and continuity of the event, a weighted statistical method can also be used, assigning different weights to different frames based on the event duration, target approach speed, and trajectory stability, thus obtaining a weighted blind zone probability. In the formula, P represents the blind zone probability, M represents the number of valid frames or valid time slices where the blind zone event occurs, and N represents the total number of frames or time slices within the preset time window. The larger this ratio, the higher the probability of a dynamic target existing in the blind zone within that time window, and the higher the corresponding blind zone danger level.

[0074] As another example, a preset time period can be divided into multiple consecutive time intervals. For instance, a 5-second time window can be divided into five 1-second sub-intervals. Within each time interval, the occurrence of blind spot events, the number of times dynamic targets appear, their duration, and their direction of movement can be statistically analyzed. The statistical results from each time interval are then summarized to obtain the vehicle's blind spot probability. Using interval-based statistics allows for the preservation of risk evidence from preceding and following time intervals even when the target is briefly occluded, the vehicle's posture changes significantly, or the image quality deteriorates instantaneously. This results in a final probability result with better continuity and resistance to interference.

[0075] Furthermore, the system can also filter the probability of blind spots, for example, by using moving average, exponential smoothing or Kalman filtering to weaken abnormal fluctuations, making the basis for subsequent early warning triggering more stable.

[0076] To ensure feasibility in this step, image preprocessing can be performed before blind zone probability detection. This includes distortion correction, brightness normalization, rain and fog enhancement, nighttime noise reduction, and backlight compensation to reduce the impact of complex weather and lighting conditions on target recognition accuracy. When the sensing hardware is not limited to cameras, millimeter-wave radar echo information, ultrasonic near-range information, and blind zone image data can be fused. The image recognition results can then be cross-validated using target distance, relative velocity, and azimuth angle to improve the accuracy of blind zone event determination.

[0077] Based on the above processing, it is clear that this step does not perform a one-time static judgment on the existence of a target within the blind zone. Instead, it transforms the frequency, duration, and trend of blind zone events within a preset time period into calculable probability information, thereby converting the originally discrete perception results into a continuous risk quantity. This effectively mitigates the problems of false alarms and missed alarms caused by rapid approaching vehicles, repeated short-term target occlusion, and single-frame false detections. It provides a stable, quantifiable, and time-continuous basic input for the generation of subsequent warning information, thereby improving the accuracy and real-time performance of risk assessment.

[0078] S102: If the blind spot probability is greater than the preset probability threshold, then target warning information is generated based on the blind spot probability and preset personalized data; personalized data represents the driver's historical driving experience; target warning information represents a risk warning prompt for blind spot hazards.

[0079] Personalized data is used to characterize a driver’s historical driving experience and may include one or more of the following: the driver’s cumulative driving mileage, driving experience, smoothness of historical lane change operations, response time after blind spot-related warnings are triggered, records of past risk events, and the receipt and handling of warning information.

[0080] Target warning information is used to characterize risk warning prompts for blind spot hazards. Its presentation can take the form of instrument panel prompts, central control display content, rearview mirror warning light status, voice announcements, buzzer alarm rhythm, steering wheel vibration signals, or a combination of these. By combining blind spot probability with personalized data for warning generation, the system outputs no longer a fixed, uniform alarm, but rather an adaptive prompt result corresponding to the current risk level and the driver's experience level.

[0081] For example, after obtaining the blind zone probability, the blind zone probability is compared with a preset probability threshold, which can be set to different levels such as 0.3, 0.5 or 0.7, to distinguish between general risk, medium risk and high risk scenarios.

[0082] When the blind spot probability exceeds a preset probability threshold, the system enters the target warning information generation process. In this process, personalized data associated with the current driver account, vehicle key identity, seat memory configuration, or facial recognition results is retrieved, and matching is performed based on the blind spot probability and the personalized data.

[0083] Based on the above analysis, it is clear that personalized data reflects the driver's historical driving experience. Therefore, the system can differentiate its warning strategies according to the driver's experience level. For example, when personalized data indicates that the driver has limited historical driving experience, a longer warning response time, or a more conservative past lane-changing maneuver, the system can generate clearer risk warning prompts, such as simultaneously illuminating the blind spot warning light, outputting a voice prompt "Vehicle approaching from the side and rear, please postpone lane changing," and increasing the volume of the buzzer or shortening the alarm interval. When personalized data indicates that the driver has extensive historical driving experience, responds promptly to warnings, and operates stably, the system can generate relatively concise risk warning prompts, such as displaying a highlighted warning icon only in the rearview mirror area or displaying a brief text prompt in the instrument panel, to avoid excessively interfering with the driver's normal operation.

[0084] As an example, the generation of target warning information also includes classifying risk warning prompts according to the magnitude of the blind zone probability. Specifically, the system can divide the blind zone probability into multiple risk level intervals, such as low-risk, medium-risk, and high-risk intervals, and map different level intervals to different prompt intensities, content, and methods.

[0085] When the blind spot probability is in the low-risk range, the target warning information can be presented as a visual reminder to prompt the driver to pay attention to the side and rear environment; when the blind spot probability enters the medium-risk range, the system adds an audible prompt to encourage the driver to pay more attention; when the blind spot probability reaches the high-risk range and the dynamic target has a rapid approaching trend, the system can output a stronger composite warning, including continuous voice, flashing icons and tactile reminders.

[0086] The tiered processing here depends not only on the absolute magnitude of the blind zone probability, but also on the increasing trend of the probability over continuous time. For example, when the blind zone probability rises rapidly in a short period of time, the warning level can be raised in advance even if the highest threshold has not yet been reached, so as to ensure the forward-looking nature of the warning timing.

[0087] To achieve the aforementioned personalized adaptation, this embodiment of the application may also pre-construct a warning strategy mapping table or train a personalized warning generation model. The warning strategy mapping table can establish the correspondence between blind spot probability intervals, driving experience levels, and output prompt modes; the personalized warning generation model can take blind spot probability, driver experience characteristics, current vehicle speed, and road scene parameters as input, and output a warning form and intensity suitable for the current driver.

[0088] Regardless of whether rule matching or model reasoning is used, the final target warning information must be directly executable by the vehicle alert system. For example, it can drive the instrument panel control module to display icons, send broadcast text to the vehicle voice module, send alarm rhythm parameters to the buzzer control module, or send vibration commands to the steering wheel actuator.

[0089] In this step, the introduction of personalized data allows for different alert strategies for the same blind spot risk among different drivers, thus solving the problem that existing solutions with uniform triggering conditions and alert content cannot adequately address driver differences. Simultaneously, blind spot probability, as a continuously quantified result, participates in warning generation, ensuring that warning triggering no longer relies on detection results at a single moment but is based on the cumulative risk over a preset time period. Based on this processing logic, the target warning information can simultaneously reflect the degree of blind spot danger and individual driver experience differences, thereby improving the targeting, reliability, and response efficiency of warning output, reducing over- or under-warning situations, and ultimately enhancing the blind spot risk assistance effect of vehicles in complex traffic environments.

[0090] Based on the above analysis, this application provides a vehicle-based risk warning method, including: detecting the blind spot probability of a vehicle, where the blind spot probability represents the probability of a blind spot event occurring within a preset time period, and the blind spot event represents an event where a dynamic target exists within the vehicle's blind spot; if the blind spot probability is greater than a preset probability threshold, then target warning information is generated based on the blind spot probability and preset personalized data, where the personalized data represents the driver's historical driving experience, and the target warning information represents a risk warning prompt for blind spot hazards. In this embodiment, by continuously collecting, statistically analyzing, and probabilistically representing the occurrence of dynamic targets within the vehicle's blind spot, the blind spot risk is elevated from a single detection result to a quantitative indicator with temporal continuity. Based on this, and combined with the driver's historical driving experience, an adaptive warning output is generated, enabling risk assessment to more accurately reflect the accumulation and changing trends of risks in complex traffic environments, while making the warning prompts more consistent with the cognitive and response characteristics of different drivers. Therefore, based on the vehicle's existing perception, processing, and prompting links, the accuracy, real-time performance, and individual adaptability of blind spot risk warnings can be improved, providing feasible technical support for safety assistance in high-speed lane changes, nighttime driving, obstructed scenarios, and complex traffic flow environments.

[0091] It should be understood that the above examples are for illustrative purposes only and are not intended to be limiting.

[0092] Figure 2 A flowchart illustrating the vehicle-based risk warning method provided in this application. Figure 2 ,like Figure 2 As shown, the above-mentioned method for detecting the blind spot probability of a vehicle includes: acquiring blind spot image data of the vehicle within a preset time period; the blind spot image data characterizing the road conditions within the vehicle's blind spot; and determining the vehicle's blind spot probability based on the blind spot image data. The above method includes:

[0093] S201. Acquire blind spot image data of the vehicle within a preset time period; the blind spot image data represents the road conditions within the vehicle's blind spot.

[0094] Blind spot image data is used to characterize road conditions in areas behind, to the sides, or other areas that are difficult for the driver to see directly. It can be continuously collected by cameras installed at the rear of the vehicle, near the exterior rearview mirrors, or on both sides of the rear of the vehicle within a preset time period.

[0095] The preset time period can be set to a continuous number of seconds based on vehicle speed, road type, or risk assessment cycle, or it can be updated at fixed intervals to ensure that the collected images can reflect the entry, exit, and continuous stay of targets within the blind spot.

[0096] To improve the effectiveness of image data, image frames can be time-synchronized, distortion-corrected, and brightness-compensated during the acquisition process, thereby reducing the impact of nighttime, backlight, or rain / fog environments on road condition recognition. In practical applications, other models of the camera device can also be selected, but this application embodiment does not limit this.

[0097] S202. Determine the vehicle's blind spot probability based on blind spot image data.

[0098] For example, when determining the probability of a vehicle's blind spot, target recognition, motion analysis, and time-series statistics can be performed on the blind spot image data to determine whether there is a dynamic target in the vehicle's blind spot, and whether the dynamic target continues to appear, appears repeatedly, or disappears briefly and then reappears within a preset time period.

[0099] If a vehicle, motorcycle, non-motorized vehicle, pedestrian, or other object that may pose a collision risk is identified, and the object appears frequently or for a long period of time in the blind spot area, the corresponding blind spot probability increases; if only occasional static background changes occur within a time period and no obvious dynamic target trajectory is formed, the corresponding blind spot probability decreases.

[0100] Furthermore, the number of targets identified in each frame of image, target category, target speed, the proportion of targets in the blind zone area, and the number of consecutive frames in which targets appear can be used as inputs for probability calculation, and combined with preset weights to generate probability values, so that the blind zone probability can more accurately reflect the possibility of blind zone events occurring.

[0101] Through the above processing, changes in the vehicle's blind spot imagery can be transformed into continuous, quantifiable probabilistic information, thus avoiding false alarms and missed alarms caused by relying solely on single frames or single target detections. This method can more comprehensively depict the changing trends of road conditions within the blind spot, and is particularly suitable for scenarios such as high-speed lane changes, rapid approach of following vehicles, and complex traffic flow interactions, providing a stable input for subsequent generation of targeted warning information based on blind spot probabilities. With this implementation, the accuracy and temporal continuity of the vehicle's blind spot hazard identification are improved, and warning triggers are closer to real-world risk changes, enhancing the driver's perception of blind spot risks and improving driving safety.

[0102] S203. If the blind zone probability is greater than the preset probability threshold, then target warning information is generated based on the blind zone probability and the preset personalized data.

[0103] In some possible implementations of this embodiment, determining the vehicle's blind spot probability based on blind spot image data includes:

[0104] Based on the blind spot image data within each preset time window, determine the marking information corresponding to each preset time window; the marking information indicates whether a blind spot event occurred within the preset time window; the preset time window is located within a preset time period; determine the vehicle's blind spot probability based on the marking information corresponding to each preset time window.

[0105] The preset time window is used to characterize statistical time units within a preset time period, which can correspond to the continuous monitoring interval during a single vehicle journey. Blind spot image data is acquired by the vehicle's rear-facing or side-rear-facing cameras and is used to characterize changes in road conditions within the vehicle's blind spot. The labeling information is used to discretely characterize whether a blind spot event occurs within a single time window. A blind spot event is used to characterize a situation where a dynamic target exists within the vehicle's blind spot and that the dynamic target meets preset hazard judgment conditions.

[0106] For example, a preset time period can be divided into multiple preset time windows of fixed duration. The windows can be connected end to end or overlap to accommodate different sampling frequencies and risk statistics requirements.

[0107] After receiving the blind spot image data in each window, the system identifies vehicles, motorcycles, pedestrians, non-motorized vehicles or their movement trajectories in the image, and determines whether a blind spot event has occurred in that window by combining the target's relative position, relative speed and duration of entry into the blind spot area.

[0108] When the determination result indicates the presence of a dynamic target meeting the criteria within the window, the corresponding marker information for that window is recorded as "Blind zone event occurred." When the determination result indicates the absence of a dynamic target meeting the criteria, the corresponding marker information for that window is recorded as "No blind zone event occurred." This marker information can be stored using binary encoding, Boolean values, or enumerated identifiers for easy subsequent statistical analysis.

[0109] After obtaining the marking information corresponding to each preset time window, the marking results of multiple windows are summarized, and the number of windows where "blind spot events occurred," the duration of consecutive occurrences, or the frequency of occurrence are statistically analyzed. These statistical results are then mapped to the vehicle's blind spot probability. This blind spot probability can be expressed as a normalized ratio, a weighted average, or an output value based on a probability model. The more frequently events occur and the longer their duration within a window, the higher the corresponding blind spot probability. In this way, discrete window determination results can be converted into a continuously comparable risk representation.

[0110] This processing mechanism achieves continuous statistical and quantitative expression of blind zone risk by performing windowed analysis over a preset time period, avoiding occasional errors caused by making judgments based solely on a single frame image. Since the blind zone probability is derived from the combined results of multiple time windows, it can more stably reflect the continuous occurrence and changing trends of dynamic targets within the blind zone, thereby improving the accuracy and real-time performance of blind zone risk assessment and providing a reliable basis for generating targeted risk warning information.

[0111] In some specific implementations of this embodiment, the method further includes: acquiring remote data; the remote data representing the road conditions ahead of the vehicle; the distance between the vehicle and the road ahead being greater than a preset distance threshold; determining a first risk state based on the remote data; the first risk state representing whether there is a driving risk on the road ahead; determining a second risk state based on the blind spot probability; the second risk state representing the change in the vehicle's blind spot probability relative to the historical blind spot probability; if the first risk state represents a driving risk on the road ahead, and the second risk state represents that the vehicle's blind spot probability is greater than the historical blind spot probability, then generating strategy information based on the first risk state and the second risk state; the strategy information representing the vehicle's risk avoidance strategy.

[0112] The remote data is used to characterize road conditions ahead of the vehicle. It can be acquired by the vehicle-to-everything (V2X) communication unit, navigation terminal, or roadside sensing unit, and is used to reflect whether there are conditions affecting traffic flow such as congestion, construction, accidents, water accumulation, icing, or sharp bends ahead. Blind spot image data is used to characterize dynamic targets and environmental changes within the vehicle's blind spots, and can come from side and rear cameras, rearview mirror cameras, or other onboard vision sensors.

[0113] When the distance between the vehicle and the road ahead is greater than a preset threshold, it means that the vehicle has not yet entered the area where close-range operation requires relying on details of the road ahead. At this time, remote data can be acquired simultaneously and analyzed together with blind spot image data.

[0114] For example, a pre-set data analysis tool is invoked. This tool could be a trained artificial intelligence model or a pre-set algorithm to analyze remote data, thereby obtaining a first risk status. The first risk status indicates whether there is a driving risk on the road ahead. For instance, by analyzing remote data, it can be determined whether there is congestion on the road ahead, whether there is a sharp turn, or whether the road ahead becomes bumpy.

[0115] Data analysis of blind spot probabilities yields a second risk state. This second risk state characterizes how the vehicle's blind spot probability changes relative to historical blind spot probabilities. For example, comparing the newly determined blind spot probability with historical blind spot probabilities determined one or more times prior to the current moment determines whether the blind spot probability has increased or decreased.

[0116] If the first risk state indicates a driving risk on the road ahead (e.g., a sharp turn ahead), and the second risk state indicates that the vehicle's blind spot probability is greater than the historical blind spot probability (i.e., the blind spot probability increases), then a pre-defined data analysis tool is invoked to process the data from both the first and second risk states, generating strategy information. This generated strategy information can provide risk avoidance solutions for both the first and second risks. The pre-defined data analysis tool can provide targeted risk avoidance solutions for specific risk states through rule matching, state machine determination, or fusion models.

[0117] For example, strategy information is used to characterize a vehicle's risk avoidance strategies, which may include deceleration and maintenance, early lane change, canceling lane change, lane keeping, limiting acceleration, or prompting the driver to take over control.

[0118] In this way, the vehicle can make a comprehensive judgment by acquiring information about the far end of the road ahead and combining it with the image of the blind spot at the near end. In this way, when there is a need to avoid danger, the vehicle can output strategy information that matches the current environment, improve the timeliness, accuracy and adaptability of the avoidance decision, and reduce the driving risks caused by the combined effect of blind spot targets and road conditions ahead.

[0119] In some specific implementations of this embodiment, the above-mentioned generation of target warning information based on blind spot probability and preset personalized data includes: determining experience information based on personalized data; experience information characterizing the driver's proficiency in driving the vehicle; determining the risk level based on blind spot probability and experience information; and generating target warning information based on the risk level.

[0120] Personalized data may include information such as the driver's years of driving experience, cumulative mileage, historical lane change frequency, blind spot warning response records, and driving habit profile. This data is normalized and mapped to experience information. Experience information is used to characterize the driver's proficiency in vehicle control, lane change judgment, and rear target perception, and can reflect the driver's ability to accept and respond to blind spot risks.

[0121] When a driver drives steadily for a long time and responds quickly to warnings, the experience information corresponds to a higher level of proficiency. When a driver has less driving experience or a history of misjudgments, the experience information corresponds to a lower level of proficiency.

[0122] The risk level can be calculated by combining blind spot probability and experience information. Based on the range of the blind spot probability and the level of the experience information, the current risk level is determined to be low, medium, or high, and warnings of different intensities are output accordingly. For example, the higher the blind spot probability and the lower the driver's skill level indicated by the experience information, the higher the risk level; conversely, the lower the blind spot probability and the higher the driver's skill level indicated by the experience information, the lower the risk level.

[0123] Target warning information can consist of one or more of the following: instrument panel prompts, rearview mirror warning lights, central control icons, beeps, or voice announcements. When the current risk level is high, the target warning information can use highlighted displays, continuous beeps, or clear lane change restriction prompts to enhance the driver's perception of blind spot hazards; when the current risk level is low, the target warning information can use a gentler reminder to avoid excessive interference to more experienced drivers. By linking blind spot probability with experience information, the system can generate differentiated warning outputs for different drivers, matching the warning content with their driving proficiency.

[0124] In these implementations, after obtaining the blind spot probability, personalized data is first converted into empirical information, and then the two are combined to determine the risk level. Finally, target warning information is generated according to the risk level, so that the warning result not only reflects the probability of blind spot events occurring, but also takes into account the driver's ability to cope with risks. This processing method can improve the pertinence and acceptability of warnings, reduce the false alarms caused by standardized prompts, and enhance the driver's response efficiency to warning information. Since the risk level is determined by both probability and experience, it can more accurately express the actual degree of danger in complex traffic environments and improve the overall effectiveness of vehicle blind spot safety assistance.

[0125] In some specific implementations of this embodiment, the above-mentioned generation of target warning information based on risk level includes: generating initial prompt information based on risk level; the initial prompt information characterizes the initially generated risk warning information for blind spot hazards; detecting the driver's emotional state; and adjusting the initial prompt information based on the emotional state to obtain the target warning information.

[0126] The initial prompt information is used to carry the early warning content pre-formed based on the risk level. The content can correspond to image prompts, sound prompts, vibration prompts or a combination of prompts, and the information intensity is matched with the risk level so that high risk corresponds to more eye-catching prompts and low risk corresponds to more concise prompts.

[0127] The driver's emotional state is used to characterize the driver's current emotional features, such as tension, anxiety, calmness, or irritability. It is typically determined by in-vehicle cameras capturing facial expressions, eye movements, and head posture, or by combining steering wheel operations, braking frequency, and voice characteristics. The emotional state recognition results can be used by the executing entity to generate state labels and output to the prompt generation module for adaptation correction of the initial prompt information.

[0128] For example, after obtaining the risk level, the prompt template library is first called to generate the initial prompt information corresponding to the risk level. The prompt template library can pre-store warning texts, icons and audio clips with different semantic strengths, such as "Caution: Vehicle approaching from the right rear" and "Blind spot: Target rapidly approaching".

[0129] Subsequently, the emotion detection module analyzes the driver's facial expression features, pupil dilation and constriction features, and voice response features to obtain emotion state parameters, which are then compared with the warning intensity threshold. When the driver is detected to be in a tense state, the originally strong warning content can be adjusted to a target warning message with clearer semantics, shorter playback duration, and more stable volume to reduce additional stimulation; when the driver is detected to be in a calm state, the original expression of the initial warning message is maintained or only slightly adjusted.

[0130] Adjustments may include word replacement of the prompt text, modification of the rhythm of the prompt audio, adjustment of the color contrast of the prompt icon, and reset of the prompt duration. All of these adjustments are completed locally in real time by the vehicle software. In practical applications, other models of this component may also be selected, but this application embodiment does not limit this.

[0131] In these implementations, initial warning information corresponding to the risk level is generated first, and then modified based on the driver's emotional state. This ensures that the output target warning information maintains consistency with the degree of danger in the blind spot while also taking into account the driver's current comprehension ability, thereby improving the readability and response efficiency of the warning information. Because the warning content can adapt to changes in the driver's emotions, it reduces the risk of misjudgment and neglect under emotional stress, and enhances the driver's attention to and adherence to blind spot risk warnings, thus improving the overall safety effect of vehicle blind spot warnings.

[0132] In some specific implementations of this embodiment, generating initial prompt information based on risk level includes: inputting the risk level into a preset model to obtain the output initial prompt information; the preset model is a trained artificial intelligence model used to generate initial prompt information based on the user's daily conversation content and risk level.

[0133] The pre-defined model can be deployed in an in-vehicle terminal, a local computing unit, or an edge computing unit connected to the vehicle. Internally, it may include a language feature encoding module, a risk level mapping module, and a prompt text generation module. During the training phase, the model receives users' everyday conversational texts, tone features, and common risk prompt samples as training data. Through supervised learning or sequence generation learning, it establishes a mapping relationship from risk levels to prompt texts, enabling the model to output prompt information that aligns with the user's understanding, based on everyday conversational expression habits.

[0134] The generated risk level is input into the preset model, which then generates initial prompts that match the current risk level. These initial prompts can be text prompts, voice prompts, or prompts displayed on the interface.

[0135] For example, after receiving the risk level, the model first vectorizes and encodes the risk level, and then combines it with the pre-learned dialogue semantic representation to generate prompts that are adapted to the driving scenario.

[0136] Because the model has learned from users' daily conversations, its initial output prompts maintain a natural style and align with users' everyday understanding, thus avoiding the limitation of fixed templates in risk warnings. If the input risk level changes, the model adjusts the intensity, length, and tone of its output prompts to ensure the initial prompts remain consistent with the current risk level.

[0137] In practical applications, the model can also be implemented using different neural network structures, such as recurrent neural networks, Transformer networks, or combinations thereof. This application embodiment does not limit this. The specific model or network size used can be configured according to the vehicle's computing power, storage resources, and real-time requirements. In practical applications, other models of this component can also be selected. This application embodiment does not limit this.

[0138] The model works by taking the risk level as input and using its language generation capabilities, learned from users' daily conversations, to output targeted initial prompts. These prompts are then passed to subsequent modules for emotion adaptation or interface presentation. Because the prompts are not fixed but dynamically generated by the model based on the risk level, the flexibility and relevance of the prompts are improved, and the stiffness caused by templated prompts is reduced.

[0139] By adopting the above method, the initial prompts can establish a stable correspondence with the risk level, and by learning from users' daily conversations, they can form more natural expressions, thereby improving the comprehensibility and acceptability of the prompts. At the same time, the model-based generation method can adapt to the differences in prompts under different risk intensities, enhancing the continuity and consistency of the system output, making vehicle risk warning information easier for drivers to identify and respond to in a timely manner.

[0140] In some specific implementations of this embodiment, generating initial prompt information based on risk level includes: obtaining customized information; the customized information represents the perception parameters of voice interaction preset by the driver; and generating initial prompt information based on risk level and customized information.

[0141] Customized information is used to characterize the driver's personalized settings for voice interaction. This information can be pre-recorded and saved by the driver in the vehicle's infotainment system, mobile application, or in-vehicle settings interface. Perception parameters may include voice prompt volume, speech rate, timbre, playback length, reminder wording, and wake-up method, reflecting the driver's perception habits and acceptance preferences for voice alerts.

[0142] Upon receiving the risk level, the system reads the corresponding customized information from the local storage unit or user account configuration file, and then matches the risk level with the customized information to determine the initial prompt information appropriate to the current risk level.

[0143] For example, when the risk level is high, the system determines whether the driver prefers direct, concise, and clear voice communication based on the customized information. If so, it generates a clear prompt containing the risk object, risk direction, and suggested action, and outputs it at a higher volume and faster response time. If the customized information indicates that the driver prefers a gentler prompt, it generates a prompt with a relatively calm tone and a relatively tactful expression, and controls the broadcast volume and speed to be within the driver's preset range.

[0144] It can also adapt the playback order, repetition frequency, and pause duration of the prompts based on customized information, so that the initial prompts can reflect the current risk level and conform to the driver's pre-set voice interaction perception parameters.

[0145] This ensures that the initial warning messages are consistent with driver preferences in both content intensity and expression. In this way, the system can output more targeted voice warnings when facing different levels of risk, reducing the inadequacy of a standardized broadcast, improving driver acceptance and response efficiency of warning information, and enhancing the practical effectiveness of blind spot risk warnings.

[0146] In some specific implementations of this embodiment, if the blind spot probability is greater than a preset probability threshold, an adjustment command is generated based on the blind spot probability; the adjustment command is used to adjust the driving environment inside the vehicle.

[0147] The control commands are environmental control signals generated based on the currently detected blind spot probability of the vehicle. These signals are used to activate the vehicle's air conditioning, audio system, seats, windows, and warning devices to change the driver's in-vehicle environment. When the blind spot probability exceeds a preset probability threshold, it indicates that the potential risk behind or to the side of the vehicle has increased to the point where increased driver attention is required. In this case, the controller can output control commands that match the risk level, changing the in-vehicle temperature, airflow intensity, warning volume, warning frequency, or ambient lighting display to assist the driver in timely risk perception.

[0148] Control commands can be generated by the vehicle controller based on blind spot probability, or they can be generated based on blind spot probability in combination with environmental control rules. Environmental control rules can be stored in the vehicle memory and establish communication connections with the air conditioning control module, vehicle multimedia module, seat control module and body control module.

[0149] For example, after receiving the blind spot probability, it is first determined whether it is greater than a preset probability threshold, and then the corresponding adjustment intensity and adjustment method are determined. For example, if the blind spot probability is greater than the preset probability threshold, based on the fact that the blind spot probability is large, i.e. the risk is high, the intensity of the warning sound is increased, the noise interference in the vehicle is reduced, or the visual cues are enhanced, and the adjustment state is maintained when the risk continues to exist.

[0150] In these implementations, by triggering the in-vehicle environment adjustment function when the blind spot probability exceeds a preset probability threshold, the vehicle no longer relies solely on single audio and visual cues. Instead, it enhances the driver's alertness and risk perception by altering the in-vehicle driving environment. Since the in-vehicle environment adjustment establishes a direct correlation with the blind spot risk state, it can enhance the warning effect and reduce the probability of missed perception in scenarios where rear targets are rapidly approaching, complex road conditions are highly disruptive, or the driver's attention is easily distracted. This improves the effectiveness of blind spot warnings and overall vehicle driving safety.

[0151] This application embodiment can be applied to nighttime driving scenarios on highways. For example, a sports utility vehicle (SMU) is continuously traveling at 100 km / h on a highway with insufficient lighting around the vehicle, making it difficult for the driver to directly observe the blind spots in the rearview mirrors. During vehicle operation, the system continuously acquires blind spot image data within a preset time period. This blind spot image data represents the real-time road conditions within the vehicle's blind spot, and the preset time period is divided into multiple preset time windows, for example, one time window every 3 seconds. The system analyzes the blind spot image data within each preset time window to determine whether there are dynamic targets in the vehicle's blind spot within the corresponding time window, thereby determining the corresponding marker information for each preset time window. The marker information is used to indicate whether a blind spot event has occurred within the preset time window. A blind spot event is an event in which a dynamic target exists within the vehicle's blind spot. In a nighttime highway scenario, a dynamic target can be a vehicle rapidly approaching from behind, or a pedestrian or other moving object entering the blind spot area.

[0152] In one possible implementation, the system can perform temporal analysis on blind spot image data within preset time windows based on continuous images captured by the rearview mirror camera. For example, it can use convolutional neural networks to identify features such as vehicle outlines and relative positions, and use long short-term memory networks to model the movement trajectory in consecutive frames to determine whether a dynamic target continuously approaching the vehicle exists within a given time window. When a vehicle is detected approaching at a high relative speed within multiple consecutive time windows, the corresponding time window is marked as a blind spot event. Subsequently, based on the marking information corresponding to each preset time window, the occurrence of blind spot events within a preset time period is statistically analyzed to determine the vehicle's blind spot probability. The blind spot probability characterizes the probability of a blind spot event occurring within that preset time period. For example, if a vehicle is repeatedly detected approaching from behind within several consecutive 3-second time windows, the blind spot probability for that preset time period can be considered high.

[0153] After determining the blind spot probability, the system compares it with a preset probability threshold. When the blind spot probability exceeds the threshold, it retrieves preset personalized data. This personalized data represents the driver's historical driving experience, and experience information is determined based on it. This experience information represents the driver's proficiency in driving. For example, personalized data may include driving experience, accident history, and historical risk response behavior. If historical records show the driver is a novice or has limited experience handling high-speed nighttime scenarios, the corresponding experience information can be determined as low proficiency. The system combines the blind spot probability and experience information to determine the current risk level. If a vehicle is rapidly approaching from behind and the driver has limited experience, the current risk level can be classified as high, and an initial warning message can be generated based on this risk level.

[0154] Initial prompts can be generated by inputting the risk level into a trained AI model, which then learns from the user's daily conversations to output appropriate initial prompts. For example, the AI ​​model could combine the driver's preferred concise expressions to generate a voice prompt such as "A vehicle is rapidly approaching from behind; please slow down immediately and stay in your lane." Alternatively, initial prompts can be generated by first acquiring pre-set voice interaction perception parameters from the driver as customized information, and then generating initial prompts based on the risk level and customized information. Voice interaction perception parameters can include speech rate, volume, prompt length, and prompt style. After initial prompts are generated, the system detects the driver's emotional state and adjusts the initial prompts accordingly to obtain target warning information. This target warning information represents a risk warning for blind spot hazards. For example, if the in-vehicle camera and voice interaction module detect that the driver is tense in foggy conditions, the system can add a calmer tone to the original instructions.

[0155] In this nighttime highway scenario, target warning information can be output through various in-vehicle interaction methods. The system can use text-to-speech to read out the warning content adjusted for the driver's emotional state, and simultaneously display the text information on the central control screen. Simultaneously, it can link with the haptic feedback module to output steering wheel and seat vibrations based on the risk level and the driver's emotional state; for example, the steering wheel vibrates at a 5 Hz frequency, and the seat vibrates at a 3 Hz intensity rhythm to provide a reminder, thus complementing the delivery of target warning information. Haptic feedback can serve as one of the forms of delivering target warning information, forming an in-vehicle risk warning output together with voice and screen display.

[0156] During vehicle operation, the system also acquires remote data, which represents the road conditions ahead of the vehicle, and the distance between the vehicle and the road ahead is greater than a preset threshold. In the example of the aforementioned SUV driving at high speed at night, the remote data can be provided by the vehicle gateway, cloud platform, or edge computing node, and includes information such as fog ahead, accident location, and visibility level, with the corresponding road location being relatively far ahead of the vehicle. When the remote data determines that there is a driving risk on the road ahead, and the blind spot image data indicates an increased probability of blind spots, the system combines the remote data and blind spot image data to generate strategy information, which represents the vehicle's risk avoidance strategy. For example, if a vehicle is detected rapidly approaching from behind, an accident is detected ahead, and fog causes extremely low visibility, the system generates strategy information such as slowing down, maintaining the current lane, moving to the right when appropriate, avoiding the accident section in advance, or guiding to a nearby service area.

[0157] Simultaneously, when the blind spot probability exceeds a threshold, a control command is generated, and the in-vehicle driving environment is adjusted accordingly. If the driver's emotional state is identified as tense and the blind spot risk persists, the control command can activate the cabin environment control system, coordinating the control of the air conditioning temperature, ventilation mode, oxygen supply module, and in-vehicle audio system. For example, it can appropriately lower the air conditioning temperature, increase oxygen supply, and play white noise to create an in-vehicle driving environment that aligns with the target warning information. As new blind spot image data, remote data, and vehicle status data are continuously acquired during subsequent driving, the system can update the blind spot probability, risk level, target warning information, strategy information, and driving environment adjustment status accordingly.

[0158] If the vehicle supports autonomous driving, it can automatically invoke the authorization authentication function when the risk level is determined to be higher than a preset risk threshold. This function authenticates the driver's driving control authority and determines whether the vehicle meets the conditions for automatic takeover. If it is determined that the driver has driving control authority and the vehicle meets the conditions for automatic takeover, intelligent driving assistance will be activated immediately to adjust the vehicle to a safe driving state.

[0159] Figure 3 The structural schematic diagram of the vehicle-based risk warning device provided in this application is as follows: Figure 3 As shown, the vehicle-based risk warning device 30 provided in this embodiment includes:

[0160] The detection module 301 is used to detect the blind spot probability of the vehicle; the blind spot probability represents the probability of a blind spot event occurring within a preset time period; the blind spot event represents an event in which a dynamic target exists in the blind spot of the vehicle.

[0161] The generation module 302 is used to generate target warning information based on the blind spot probability and preset personalized data if the blind spot probability is greater than a preset probability threshold; the personalized data represents the driver's historical driving experience; the target warning information represents a risk warning prompt for blind spot hazards.

[0162] In one possible implementation, the detection module 301 is further configured to acquire blind spot image data of the vehicle within a preset time period; the blind spot image data characterizes the road conditions within the vehicle's blind spot.

[0163] The probability of a vehicle's blind spot is determined based on blind spot image data.

[0164] In one possible implementation, the detection module 301 is further configured to determine the marking information corresponding to each preset time window based on the blind spot image data within each preset time window; the marking information indicates whether a blind spot event has occurred within the preset time window; the preset time window is located within a preset time period.

[0165] The probability of a vehicle's blind spot is determined based on the marker information corresponding to each preset time window.

[0166] In one possible implementation, the vehicle-based risk warning device 30 further includes a strategy module for acquiring remote data; the remote data characterizes the road conditions ahead of the vehicle; the distance between the vehicle and the road ahead is greater than a preset distance threshold.

[0167] Based on remote data, the first risk status is determined; the first risk status indicates whether there is a driving risk on the road ahead.

[0168] The second risk state is determined based on the blind spot probability; the second risk state characterizes the change in the vehicle's blind spot probability relative to the historical blind spot probability.

[0169] If the first risk state indicates a driving risk on the road ahead, and the second risk state indicates that the vehicle's blind spot probability is greater than the historical blind spot probability, then strategy information is generated based on the first and second risk states; the strategy information represents the vehicle's risk avoidance strategy.

[0170] In one possible implementation, the generation module 302 is further configured to determine experience information based on personalized data; the experience information characterizes the driver's proficiency in driving the vehicle.

[0171] The risk level is determined based on the probability of blind spots and empirical information;

[0172] Based on the risk level, generate target early warning information.

[0173] In one possible implementation, the generation module 302 is further configured to generate initial warning information based on the risk level; the initial warning information represents the preliminary risk warning information generated for blind spot hazards;

[0174] Detect the driver's emotional state;

[0175] Based on the emotional state, the initial prompts are adjusted to obtain target warning information.

[0176] In one possible implementation, the generation module 302 is further configured to input the risk level into a preset model to obtain the output initial prompt information; the preset model is a trained artificial intelligence model, which is used to generate initial prompt information based on learning from the user's daily conversation content and according to the risk level.

[0177] In one possible implementation, the generation module 302 is further configured to acquire customized information; the customized information represents the perceptual parameters of the voice interaction preset by the driver.

[0178] Generate initial alert information based on risk level and customized information.

[0179] In one possible implementation, the vehicle-based risk warning device 30 further includes a scheduling module, which generates a control command based on the blind spot probability if the blind spot probability is greater than a preset probability threshold; the control command is used to adjust the driving environment inside the vehicle.

[0180] The vehicle-based risk warning device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0181] Figure 4 This is a structural diagram of the vehicle-based risk warning device provided in this application. Figure 4 As shown, the vehicle-based risk warning device 40 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the vehicle-based risk warning device 40 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0182] In the specific implementation process, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to execute the above-mentioned vehicle-based risk warning method.

[0183] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0184] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0185] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0186] 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.

[0187] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described vehicle-based risk warning method.

[0188] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the aforementioned vehicle-based risk warning method.

[0189] The aforementioned readable 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 readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0190] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0191] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components 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 units, and may be electrical, mechanical, or other forms.

[0192] The units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0193] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0194] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0195] 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.

[0196] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A vehicle-based risk warning method, characterized in that, include: Detect the probability of blind spots in a vehicle; The blind zone probability represents the probability of a blind zone event occurring within a preset time period; The blind spot event represents an event in which a dynamic target exists within the vehicle's blind spot; If the blind zone probability is greater than a preset probability threshold, then target warning information is generated based on the blind zone probability and preset personalized data. The personalized data represents the driver's historical driving experience; the target warning information represents risk warning prompts for blind spot hazards.

2. The method according to claim 1, characterized in that, Detecting the probability of a vehicle's blind spots includes: Acquire blind spot image data of a vehicle within a preset time period; the blind spot image data represents the road conditions within the vehicle's blind spot. Based on the blind spot image data, the probability of the vehicle's blind spot is determined.

3. The method according to claim 2, characterized in that, Based on the blind spot image data, the probability of the vehicle's blind spot is determined, including: Based on the blind spot image data within each preset time window, the marking information corresponding to each preset time window is determined; the marking information indicates whether a blind spot event has occurred within the preset time window; the preset time window is located within the preset time period. The probability of a vehicle's blind spot is determined based on the marker information corresponding to each preset time window.

4. The method according to claim 1, characterized in that, The method further includes: Acquire remote data; the remote data represents the road conditions ahead of the vehicle; the distance between the vehicle and the road ahead is greater than a preset distance threshold; Based on the remote data, a first risk status is determined; the first risk status indicates whether there is a driving risk on the road ahead. Based on the blind spot probability, a second risk state is determined; the second risk state characterizes the change in the vehicle's blind spot probability relative to the historical blind spot probability. If the first risk state indicates a driving risk on the road ahead, and the second risk state indicates that the vehicle's blind spot probability is greater than the historical blind spot probability, then strategy information is generated based on the first risk state and the second risk state; the strategy information represents the vehicle's risk avoidance strategy.

5. The method according to claim 1, characterized in that, Based on the blind zone probability and preset personalized data, target early warning information is generated, including: Based on the personalized data, experience information is determined; the experience information represents the driver's proficiency in driving the vehicle. The risk level is determined based on the blind spot probability and the empirical information. Based on the risk level, generate target early warning information.

6. The method according to claim 5, characterized in that, Based on the risk level, target early warning information is generated, including: Based on the risk level, an initial warning message is generated; the initial warning message represents a preliminary risk warning message for blind spot hazards. Detect the driver's emotional state; Based on the emotional state, the initial prompt information is adjusted to obtain the target warning information.

7. The method according to claim 6, characterized in that, Based on the risk level, an initial warning message is generated, including: The risk level is input into a preset model to obtain the initial prompt information output; the preset model is a trained artificial intelligence model used to generate initial prompt information based on learning from the user's daily conversation content and according to the risk level.

8. The method according to claim 6, characterized in that, Based on the risk level, an initial warning message is generated, including: Obtain customized information; the customized information represents the perception parameters of the voice interaction preset by the driver. Based on the risk level and the customized information, an initial prompt message is generated.

9. The method according to any one of claims 1-8, characterized in that, If the blind spot probability is greater than a preset probability threshold, an adjustment command is generated based on the blind spot probability; the adjustment command is used to adjust the driving environment inside the vehicle.

10. A vehicle-based risk warning device, characterized in that, include: The detection module is used to detect the probability of blind spots in a vehicle. The blind zone probability represents the probability of a blind zone event occurring within a preset time period; The blind spot event represents an event in which a dynamic target exists within the vehicle's blind spot; The generation module is used to generate target warning information based on the blind zone probability and preset personalized data if the blind zone probability is greater than a preset probability threshold. The personalized data represents the driver's historical driving experience; the target warning information represents risk warning prompts for blind spot hazards.

11. A vehicle-based risk warning device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-9.

12. 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 1-9.

13. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-9.

Images