Blind zone early warning method and device and storage medium

Abstract

The invention provides a blind area early warning method and device and a storage medium, and the method comprises the steps: carrying out the multi-dimensional recognition of a vehicle road scene, obtaining the data of a plurality of sensors, determining the environment type of the road scene where a vehicle is located, adjusting the original weight of each sensor based on the environment type, obtaining a first weight corresponding to each sensor, and carrying out the adjustment of the first weight. Through recognition of environment types, weights are dynamically distributed based on indexes related to environment scenes in a non-linear mode so as to adapt to complex scenes. The sensor data is fused according to the first weight, the influence of the data source with low confidence on the fusion processing is dynamically reduced, blind area early warning is performed based on the fusion result, and the false alarm rate is effectively reduced. In addition, according to dynamic weight adjustment of the indexes related to the environment, the response speed of system execution control is increased.

Description

Technical Field

[0001] This manual relates to the field of intelligent driving technology, and in particular to blind spot warning methods, devices and storage media. Background Technology

[0002] In the intelligent driving technology system, the blind spot warning system, as a key component of active safety protection, uses multi-sensor data fusion technology to accurately detect and identify targets in the vehicle's blind spots, providing drivers with reliable information for lane-changing decisions. Currently, this system mainly integrates heterogeneous data from multiple sources, including visual sensors (such as cameras), millimeter-wave radar, lidar, and high-precision maps. It uses fixed-weight or linear weighted algorithms for data fusion processing and provides warnings based on the fused results.

[0003] However, existing blind spot warning technologies have significant shortcomings in adaptability to complex environments. In adverse weather conditions such as rain, fog, and sandstorms, visual sensors are affected by environmental factors, resulting in a severe decline in image quality and a significant decrease in the confidence level of target detection. However, traditional fixed-weight or linearly weighted data fusion algorithms still assign high raw weights to the visual sensor data. This makes it difficult for the system to accurately identify targets in blind spots when the reliability of visual sensor data is reduced, easily leading to false alarms or missed alarms. This seriously affects the reliability and stability of the blind spot warning system and fails to meet the high standards of driving safety required in intelligent driving scenarios. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this specification provides a blind spot early warning method, device and storage medium.

[0005] According to a first aspect of the embodiments of this specification, a blind spot warning method is provided, the method comprising: Multi-dimensional identification of vehicle road scenes yields data from multiple sensors; Determine the environmental type of the road scene where the vehicle is located; The original weights of each sensor are adjusted based on the environment type to obtain the first weights corresponding to each sensor. The sensor data is fused according to the first weight, and blind spot warning is performed based on the fusion result.

[0006] According to a blind spot warning method provided in this specification, adjusting the original weights of each sensor based on the environment type to obtain a first weight corresponding to each sensor includes: Based on the environment type, a preset exponent value for nonlinear operation is determined; the nonlinear operation is used to adjust the original weights to match the nonlinear characteristics of the sensor under different road scenarios. The sensor also includes a confidence level. Based on the index value and the confidence level of the sensor, a nonlinear calculation is performed to obtain a first weight corresponding to each of the sensors.

[0007] According to the blind spot warning method provided in this specification, in the nonlinear operation, the confidence level is positively correlated with the first weight.

[0008] According to the blind spot warning method provided in this specification, before adjusting the original weights of each sensor based on the environment type, the method further includes: Environmental data related to environmental identification is obtained from the sensor data; Based on the environmental data and the corresponding second weight, the environmental type of the road scene where the vehicle is located is identified to obtain the initial environmental type; When the confidence level of the initial environment type is greater than or equal to the confidence threshold, the initial environment type is determined to be the final environment type of the road scene where the vehicle is located.

[0009] According to the blind spot warning method provided in this specification, the method further includes: When the confidence level of the initial environment type is less than a set threshold, the sensor whose confidence level is less than the abnormal threshold is determined to be an abnormal sensor; The second weight of the abnormal sensor is allocated to the backup sensor according to a set transfer ratio, and the third weight of the abnormal sensor after allocation and the fourth weight of the backup sensor after allocation are determined. Based on the environmental data and third weight of the abnormal sensor, the environmental data and fourth weight of the backup sensor, and the environmental data and second weight of other sensors, the environmental type of the road scene where the vehicle is located is identified, and the final environmental type of the road scene where the vehicle is located is obtained.

[0010] According to the blind spot warning method provided in this specification, the method further includes: When the confidence level of the abnormal sensor meets the weight recovery condition, the third weight of the abnormal sensor is updated to the second weight before the abnormal sensor was assigned. The weight recovery condition includes: the confidence level of the abnormal sensor remains greater than the abnormal threshold for a set period of time.

[0011] According to a blind zone early warning method provided in this specification, the confidence level of the initial environment type is obtained in the following way: The confidence levels of each environmental data point are weighted and fused to determine the confidence level of the initial environmental type.

[0012] According to the blind spot warning method provided in this specification, the step of fusing sensor data according to a first weight and performing blind spot warning based on the fusion result includes: The sensor data is fused according to the first weight to obtain a fusion result indicating the road scene within the blind spot of the vehicle; Based on the fusion results, an early warning control strategy is selected from multiple preset hierarchical early warning control strategies, and the working state of the blind spot early warning function is dynamically scheduled according to the selected early warning control strategy. When the blind spot early warning function is activated, risk monitoring is performed continuously.

[0013] According to the blind spot warning method provided in this specification, after fusing the sensor data according to the first weight and performing blind spot warning based on the fusion result, the method further includes: Obtain the results of early warning and intervention; The exponential value of the preset nonlinear operation is optimized based on the early warning intervention result; the nonlinear operation is used to adjust the original weights to match the nonlinear characteristics of the sensor under different road scenarios; During a vehicle driving cycle, and when driving under the same environmental type, the original weights of each sensor are adjusted based on the optimized index value to obtain a first weight corresponding to each sensor.

[0014] According to a second aspect of the embodiments of this specification, an apparatus is provided, comprising: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements any of the blind spot warning methods described above.

[0015] According to a third aspect of the embodiments of this specification, a computer-readable storage medium is provided, comprising: When the computer program is executed by the processor, it implements any of the blind spot warning methods described above.

[0016] The technical solutions provided in the embodiments of this specification may include the following beneficial effects: In this embodiment, multi-dimensional identification of the vehicle road scene is performed to obtain data from multiple sensors, determining the environmental type of the road scene where the vehicle is located. Based on the environmental type, the original weights of each sensor are adjusted to obtain a first weight corresponding to each sensor. Through environmental type identification, weights are dynamically allocated non-linearly based on an environment-related index to adapt to complex scenarios. Sensor data is fused according to the first weight, dynamically reducing the impact of low-confidence data sources on the fusion process. Blind spot warnings are performed based on the fusion results, effectively reducing the false alarm rate. Furthermore, the dynamic weight adjustment based on environment-related indices improves the response speed of the system's execution control.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this specification. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this specification and, together with the description, serve to explain the principles of this specification.

[0019] Figure 1 This specification illustrates a blind spot warning system architecture diagram based on an exemplary embodiment.

[0020] Figure 2 This is a flowchart illustrating a blind spot warning method according to an exemplary embodiment of this specification.

[0021] Figure 3 This is a flowchart illustrating an environmental classification module according to an exemplary embodiment of this specification.

[0022] Figure 4 This is a weight transfer timing diagram illustrating a redundant arbitration mechanism according to an exemplary embodiment of this specification.

[0023] Figure 5 This is a nonlinear weighting curve illustrated in this specification according to an exemplary embodiment.

[0024] Figure 6 This is a comparative analysis diagram of the weight allocation in a construction scenario illustrated in this specification based on an exemplary embodiment.

[0025] Figure 7 This is a block diagram illustrating an apparatus according to an exemplary embodiment of this specification.

[0026] Figure 8 This is a schematic diagram of a device illustrated in this specification according to an exemplary embodiment. Detailed Implementation

[0027] 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 numerals 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 specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0028] The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of this specification. The singular forms “a,” “the,” and “the” as used in this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0029] It should be understood that although the terms first, second, third, etc., may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this specification, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0030] This specification provides a blind spot warning method, device, and computer-readable storage medium. The embodiments of this specification are described in detail below with reference to the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] The vehicles described in this document have the conventional structure and functions of vehicles currently on the market (such as body, powertrain, chassis, electrical and electronic systems, etc.), and will not be described in detail here.

[0032] Time To Collision (TCC) is a core indicator for collision risk assessment in intelligent driving systems and a key indicator for triggering blind spot warnings. It indicates the time required for a collision to occur between the vehicle and a target object based on the current relative speed and trajectory. When the actual TCC value is less than the set warning threshold, the blind spot warning system is triggered to alert the driver of a collision risk in the current driving environment. However, significant false alarms occur in practical applications. For example, if another vehicle is traveling in the same direction in the adjacent lane, and there is a solid lane line between the two lanes, the traditional blind spot warning function will still issue a warning signal based on TCC calculations. However, in reality, the driver is aware of the solid lane line and, according to traffic rules, will not change lanes. Indiscriminately issuing a warning in this situation will cause unnecessary interference to the driver, potentially leading to misoperation due to the warning interference. Another example is when there is a guardrail on one side of the vehicle. The traditional blind spot warning function will consider avoiding a collision and issue a warning signal. Similarly, according to traffic rules, the driver is unlikely to change lanes when the guardrail is known, thus also causing false alarms. Therefore, how to make the blind spot warning function accurate based on the actual scenario is a difficult problem.

[0033] However, blind spot warning technology has significant shortcomings in adaptability to complex environments. In adverse weather conditions such as rain, fog, and sandstorms, visual sensors are affected by environmental factors, resulting in a severe decline in image quality and a significant decrease in the confidence level of target detection. However, traditional fixed-weight or linearly weighted data fusion algorithms still assign high raw weights to the visual sensor data. This makes it difficult for the system to accurately identify targets in blind spots when the reliability of visual sensor data is reduced, easily leading to false alarms or missed alarms. This seriously affects the reliability and stability of the blind spot warning system and fails to meet the high standards of driving safety required in intelligent driving scenarios. Therefore, this specification provides a solution for accurately identifying vehicle road scenes, dynamically optimizing the blind spot warning strategy based on road scenes, enabling the blind spot warning function to adapt to road scenes and provide accurate warnings, reducing false alarms and improving vehicle safety.

[0034] To address the aforementioned technical problems, this specification provides a blind spot warning method.

[0035] Aimed at real-time monitoring of sensor data quality, this system identifies environmental types and adjusts weight distribution based on scene-related indices to match the nonlinear characteristics of sensors under different road scenarios. This dynamically reduces the impact of low-confidence data sources on fusion processing, improving the accuracy of vehicle road scene recognition. Accurate scene recognition effectively reduces the system's false alarm rate. Simultaneously, dynamic weight adjustment enhances the system's response speed for control execution and effectively improves its robustness.

[0036] The following is an embodiment of a blind spot warning method provided in this specification.

[0037] like Figure 1 As shown, Figure 1 This specification illustrates a blind spot warning system architecture diagram based on an exemplary embodiment.

[0038] The blind spot early warning system comprises a data layer, a decision-making layer, and an execution layer. The decision-making layer includes a pre-decision layer and a final decision-making layer.

[0039] The data layer includes a scene recognition module, which collects multi-sensor data of road scenes. This data is used in the final decision layer to identify road scenes, determine their legality, and optimize subsequent blind spot warning functions. The multi-sensor data includes, but is not limited to, road structure data, lane line types, vehicle status parameters, and environmental types. Examples include map data and V2X-related information (including real-time V2X updates), visual perception data (including camera / LiDAR fusion data, which includes visual and point cloud data), and necessary vehicle parameter information (including steering wheel angle, yaw rate, turn signal, etc.).

[0040] The pre-decision layer pre-configures multiple tiered early warning control strategies and the mapping relationship between these strategies and road scenarios to dynamically control the current blind spot early warning function, which is crucial for achieving accurate early warning. These road scenarios include, but are not limited to, non-reversible lane scenarios, reversible lane scenarios, and transitional scenarios between the two (such as tidal flow lanes).

[0041] The final decision layer fuses multi-dimensional data to determine the legitimacy of road scenarios and selects early warning and control strategies based on this. In the process of determining the legitimacy of road scenarios, a confidence level decision is further introduced to enhance the accuracy of scenario recognition results.

[0042] The execution layer is used to implement the optimized blind spot warning function, which is achieved through the HMI and the linkage vehicle control system.

[0043] It should be noted that the functions and roles of the above layers will be explained in detail in subsequent chapters, and will not be repeated here.

[0044] like Figure 2 As shown, Figure 2 This is a flowchart illustrating a blind spot warning method according to an exemplary embodiment, comprising the following steps: In step 102, the vehicle road scene is identified in multiple dimensions to obtain data from multiple sensors.

[0045] The main drawback of traditional blind spot warning functions is the false alarm problem in scenarios where lane changes are not allowed. Therefore, when the blind spot warning function is activated, it is necessary to identify in real time whether the road scene within the vehicle's blind spot is a scenario where lane changes are not allowed, and to provide adaptive warnings in such scenarios to ensure vehicle safety while improving the driving experience.

[0046] First, the vehicle road scene is identified in multiple dimensions, and data from multiple sensors are obtained.

[0047] As an example, the multi-sensor data includes, but is not limited to, at least one of road structure data, lane line type, vehicle status parameters, visual weather, and V2X information.

[0048] The road structure data is acquired through map data (including, but not limited to, high-precision maps and standard maps) and V2X (vehicle-to-everything) wireless communication technology. The map data includes markings of all insurmountable road structures such as ramps, emergency lanes, guardrails, medians, and construction zones. The V2X wireless communication system receives dynamic traffic control information in real time, such as prohibitions on lane changes and passage. By identifying non-reversible lane areas using map data and V2X information, the system can accurately locate special road structures such as ramp merging areas and emergency lanes, enabling the identification of non-reversible lane areas outside designated lane lines and improving the accuracy of subsequent blind spot warnings.

[0049] Simultaneously, after identifying the non-changeable lane areas, these areas are marked to facilitate subsequent lane scene recognition. The marking methods may include, but are not limited to, non-changeable lane labels, highlighting, and text annotations.

[0050] Furthermore, road classification is extracted from the road structure data, including but not limited to highways and urban roads, for subsequent identification of environment types.

[0051] The lane marking type is determined through a visual perception system. This system includes multi-source sensors, including but not limited to cameras, lidar, and millimeter-wave radar. Based on data collected from the vehicle's sensors (including blind spots), the system performs target detection on the collected data to determine the lane marking type.

[0052] As an example, the sensors include a camera and a lidar, which collect data about the vehicle's driving environment, respectively, to obtain camera images and lidar point clouds. Based on object detection algorithms (such as the YOLOv8 algorithm), lane lines are classified, supporting the recognition of solid and dashed lines and tidal lane lines.

[0053] Furthermore, visual sensors are also used to identify visual weather, such as using cameras to determine the level of rain, fog, and dust storms, for subsequent identification of environmental types.

[0054] The vehicle status parameters are read in real time via the onboard CAN bus, covering dynamic data such as vehicle speed, steering wheel angle, and turn signal status. The steering wheel angle is read by a yaw rate sensor and is used to indicate the vehicle's lane-changing tendency / intention. For example, when the vehicle's real-time yaw rate exceeds a certain threshold, it is determined that the vehicle intends to change lanes. In practical applications, when the yaw rate sensor fails, the yaw angle is calculated using the steering wheel angle and vehicle speed as a backup.

[0055] Traditional methods of judging road scenarios using data from a single data source have certain limitations. For example, map data is highly reliable for static road structures (such as ramps, emergency lanes, and construction sites), but it cannot respond to dynamic changes in real time (such as temporary construction). Visual perception can capture real-time road conditions, but it is greatly affected by weather and lighting (such as reduced lane line recognition rates in rainy or foggy weather). Vehicle status is used to judge and reflect driver intentions (such as yaw rate and steering wheel angle), but it cannot independently determine the legality of the road. Therefore, a multi-dimensional scene recognition technology is proposed, which integrates map data, V2X data, visual information, and vehicle status parameters to accurately identify areas where lane changes are not allowed. This achieves a "map-vision-vehicle status closed-loop verification," solving the risk of single-point failure in current technologies.

[0056] Next, after completing the synchronous collection of multi-dimensional data through the above data sources, a three-layer data fusion processing model needs to be constructed. Through scientific weight allocation and data fusion mechanism, the real-time identification of the legality of road scenes can be achieved.

[0057] Specifically, each data source independently judges the road scene based on its own perception capabilities to form an initial recognition result.

[0058] Map data source: Based on pre-loaded road structure information and real-time matched vehicle positions, the map directly outputs the lane change legality attribute of the current road segment. For example, when a vehicle is in the ramp merging area, the map clearly marks this area as a lane change prohibited area and outputs the initial identification result of "scenario illegal" (for ease of understanding, illegal scenarios will be referred to as "lane change prohibited scenarios").

[0059] Visual perception data source: Lane line types and road signs are identified and analyzed using image recognition algorithms. If a solid lane line or a no-lane-change sign is identified, the initial recognition result for the "no-lane-change scenario" is output; if a dashed lane line is identified and there is no no-lane-change sign, the initial recognition result for the "changeable lane scenario" is output.

[0060] Vehicle status parameter data source: Determines the rationality of lane change intentions based on vehicle driving status parameters. When parameters such as vehicle speed and steering wheel angle indicate that the vehicle has an abnormal tendency to change lanes, such as suddenly turning the steering wheel sharply while driving at high speed, the initial result "Lane change scenario is questionable" is output.

[0061] Based on the initial identification results and weights corresponding to each data source, the legality of the road scene within the vehicle's blind spot is determined, indicating whether the current scene belongs to a non-changeable lane scenario. Subsequently, this result is used to select a warning control strategy from multiple preset hierarchical warning control strategies, and the working state of the blind spot warning function is dynamically scheduled according to the selected warning control strategy to realize blind spot warning.

[0062] However, the reliability of various data sources can vary depending on the vehicle's driving environment. For example, in adverse weather conditions, the recognition accuracy of visual perception data sources may decrease, thus reducing their weight; while map data sources are less affected by weather, allowing for a higher weight. However, current technologies using fixed weight allocation cannot adapt to the dynamic changes in complex environments (such as rain, fog, or construction zones), leading to high false alarm rates. Therefore, it is necessary to comprehensively consider factors such as data source reliability and environmental adaptability to dynamically determine the weights of each data source.

[0063] As an example, by dynamically allocating non-linear environmental types, the weights of the corresponding data sources are dynamically adjusted according to the environmental type of the road scene in which the vehicle is located.

[0064] Next, determine the environmental type of the road scene where the vehicle is located.

[0065] The environmental types include, but are not limited to, highway scenarios, construction scenarios, urban scenarios, tunnel scenarios, and nighttime scenarios. The data emphasis of the multi-sensor systems differs depending on the scenario type. For example, in highway scenarios, map data is more accurate and has stronger environmental adaptability, requiring a higher weight for map data. Therefore, the weights of each sensor will be dynamically adjusted according to the environmental type to match the nonlinear characteristics of sensors in different road scenarios and improve the accuracy of early warnings in complex scenarios.

[0066] In some embodiments, before adjusting the original weights of each sensor based on the environment type, the environment type of the road scene where the vehicle is located is determined; determining the environment type of the road scene where the vehicle is located includes: Environmental data related to environmental identification is obtained from the sensor data; Based on the environmental data and the corresponding second weight, the environmental type of the road scene where the vehicle is located is identified to obtain the initial environmental type; When the confidence level of the initial environment type is greater than or equal to the confidence threshold, the initial environment type is determined to be the final environment type of the road scene where the vehicle is located.

[0067] like Figure 3 As shown, Figure 3 This specification illustrates an environmental classification module flowchart based on an exemplary embodiment, in which the environmental classification module determines the environmental type of the road scene where the vehicle is located.

[0068] Data related to environmental identification is extracted from the multi-sensor data collected from the aforementioned multi-source data sources, including but not limited to map data for identifying road grades, V2X information for identifying real-time information signs, visual perception information for identifying weather grades, and information related to environmental type identification from other data sources.

[0069] Based on the environmental data extracted above for environmental identification and the second weight of each environmental data point, the initial environment type of the road scene where the current vehicle is located is determined. It should be noted that the second weight is a preset weight for each environmental data point, and different environmental data points correspond to different second weights.

[0070] Because sensor acquisition accuracy is affected by environmental factors, low acquisition accuracy results in low-reliability environment type information, which can easily lead to false alarms in subsequent warnings. Therefore, it is necessary to verify the reliability of environment type determination. Only environment type results that meet the reliability requirements will be used in the subsequent weight adjustment process.

[0071] In some implementations, the confidence levels of each of the environmental data are weighted and fused to determine the confidence level of the initial environmental type.

[0072] Use c env This represents a quantitative value indicating the reliability of the system's judgment on the current environment type, ranging from [0, 1]. When c env A value greater than or equal to 0.6 indicates high reliability in environmental classification, and the pre-defined weight allocation strategy will be applied. env <0.6: Low reliability of environmental classification, triggering a multi-redundant arbitration mechanism.

[0073] The process of determining the confidence level of the environment type of the road scene is as follows: 1) Map data confidence level c map Source: Road classification matching results from high-precision maps (such as highways and urban roads).

[0074] Setting: If the matching error between real-time positioning data and map road network is less than 1 meter, then c map = 0.9; when the error is between 1 and 3 meters, c map= 0.6.

[0075] 2) Visual weather confidence level c weather Source: Camera's recognition results of rain, fog, and dust.

[0076] Settings: No rain or fog: c_weather = 1.0.

[0077] Light rain / fog (visibility > 100 meters): c weather = 0.7.

[0078] Severe rain and fog (visibility < 50 meters): c weather = 0.3.

[0079] 3) V2X consistency verification result c v2x Source: V2X real-time traffic sign and map data conflict detection.

[0080] Settings: No conflicts (e.g., construction area displayed on map and V2X synchronized notifications): c v2x =1.0.

[0081] There is a conflict (e.g., the map is not updating the construction area but V2X is displaying a message): c v2x =0.4.

[0082] 4) Weighted fusion of the confidence levels of each environmental data point, and the comprehensive calculation formula is as follows: c env =β1*c map +β2*c weather +β3*c v2x In this process, weights are assigned based on the priority of each data source. For example, high-precision maps typically have higher information accuracy, so map data is given primary weight, followed by weather influence, with V2X verification as a supplement. That is, map data has a weight of β1=0.5, visual weather data has a weight of β2=0.3, and V2X data has a weight of β3=0.2.

[0083] It should be noted that the parameter values ​​set above are defined according to actual needs, including but not limited to needs such as energy consumption, response speed, and cost. There are no specific limitations here, and the specific values ​​are only for illustrative purposes.

[0084] If c env A value ≥ the credibility threshold indicates that the current initial environment type is credible and will be used as the final environment type of the road scene where the vehicle is located for subsequent fusion processing.

[0085] If c envA threshold below the credibility threshold indicates that the initial environment type obtained from the current environment classification has low credibility, triggering a remedial strategy. The remedial strategy includes a multi-redundancy arbitration mechanism or a dynamic weighting mechanism.

[0086] For example, if the confidence threshold is set to 0.6, when driving within a highway construction zone, if the map data has not been updated, the confidence levels for the corresponding environmental data are as follows: c map =0.6 (Construction area not marked on the map, positioning error 2 meters); c weather =0.7 (light rain / fog); c v2x =0.4 (V2X indicates a construction zone, conflicting with the map); Based on this, a weighted fusion calculation is performed on the confidence levels of the environmental data: c env =0.5×0.6+0.3×0.7+0.2×0.4=0.59 The confidence level of the initial environment type is: c env =0.59.

[0087] c env If the value is less than 0.6, a remedial strategy is triggered.

[0088] Strategy 1: Trigger dynamic weight reduction to increase the weight of V2X data.

[0089] Strategy 2: Multi-sensor redundancy arbitration mechanism. The specific process is as follows: As an example, sensors with confidence levels below an abnormal threshold are identified as abnormal sensors; The second weight of the abnormal sensor is allocated to the backup sensor according to a set transfer ratio, and the third weight of the abnormal sensor after allocation and the fourth weight of the backup sensor after allocation are determined. Based on the environmental data and third weight of the abnormal sensor, the environmental data and fourth weight of the backup sensor, and the environmental data and second weight of other sensors, the environmental type of the road scene where the vehicle is located is identified, and the final environmental type of the road scene where the vehicle is located is obtained.

[0090] The abnormal threshold refers to an abnormality in the reliability of the data source, which may be due to sensor malfunction, obstruction, or other reasons. Therefore, the corresponding sensor needs to be processed.

[0091] When the sensor confidence level c iWhen an abnormal threshold is reached, its weight is proportionally transferred to a backup sensor. It should be noted that this system supports multi-level redundancy design: it supports multiple sensors (vision / LiDAR / millimeter-wave radar, etc.) as backups for each other to avoid single-sensor failure. In the event of a sensor failure, it switches to a backup sensor, ensuring system robustness in the event of sensor failure and preventing degradation failure.

[0092] The arbitration rules are as follows: Where α is the set transfer ratio, and the adjustment range of α is [0.2, 0.4], which represents the weight ratio transferred from the abnormal sensor; The original weights for the backup sensors; This is the new weight after the backup sensor takes over, i.e., the fourth weight; The original weights of the abnormal sensors are the second weights. The weights for abnormal sensor losses; This is the new weight after the abnormal sensor weight transfer, i.e., the third weight.

[0093] By combining the weights of the environmental data from each sensor after the aforementioned redundant arbitration, and the corresponding environmental data, the environmental type of the road scene where the vehicle is located is re-identified, resulting in the final environmental type of the road scene. It should be noted that for sensors that did not exhibit anomalies, the weights of their environmental data are still calculated according to the second weight, i.e., the original weight, to comprehensively identify the environmental type of the current road scene.

[0094] like Figure 4 As shown, Figure 4 This is a weight transfer timing diagram illustrating a redundant arbitration mechanism according to an exemplary embodiment of this specification.

[0095] For example, if the anomaly threshold is set to 0.4, that is, when the sensor confidence level c... i When the value is less than 0.4, its weight is proportionally transferred to the backup sensor.

[0096] Using vision and LiDAR sensors as examples. The vision weight w is known. vision =0.4、w lidar =0.2, set the transfer ratio α=0.3.

[0097] When visual confidence level c vision When c = 0.3,vision If the value is less than 0.4, the redundant arbitration mechanism is triggered, and weight transfer calculation is performed.

[0098] Results: The weight of LiDAR was increased to 0.32, while the weight of vision was reduced to 0.28.

[0099] This means that when the camera is obscured by mud and water with a confidence level of c_vision=0.3, this solution will transfer 30% of the visual weight to the LiDAR, and the estimated system availability will remain above 95%, while the traditional system, due to the lack of weight adjustment, will have an availability of 60%.

[0100] In some embodiments, the method further includes: When the confidence level of the abnormal sensor meets the weight recovery condition, the third weight of the abnormal sensor is updated to the second weight before the abnormal sensor was assigned. The weight recovery condition includes: the confidence level of the abnormal sensor remains greater than the abnormal threshold for a set period of time.

[0101] Once a sensor has been adjusted (e.g., the obstruction is removed) and has a certain level of reliability, its weight is restored for subsequent environment type identification.

[0102] The weight recovery condition is set as: c i (t) >Abnormal threshold and ∀τ∈[t−n,t], c i (τ) >Abnormal threshold; Where τ represents a certain moment, and n is the time during which the confidence level is continuously greater than the anomaly threshold.

[0103] It's understandable that the weight recovery condition is that weight recovery is only allowed to begin when the abnormal sensor remains stable and reliable for n consecutive seconds (confidence level > abnormal threshold). Typically, the value of n is determined based on actual needs, for example, n = 5.

[0104] When the confidence level of an abnormal sensor meets the weight recovery condition, the sensor weight is restored to its original weight.

[0105] The specific sensor weight recovery process involves restoring the original weights at a rate of 0.1 per second. This prevents sudden weight changes and ensures a controllable rate of change, as detailed below: w i (t+1) =w i (t) +0.1*(w i_original -w i (t) ), where w i_original : The original weights (pre-fault values) of sensor i, wi(t) : The weight at the current time t.

[0106] The above embodiments accurately determine the environmental type of the road scene where the vehicle is located.

[0107] In step 104, the original weights of each of the sensors are adjusted based on the environment type to match the nonlinear characteristics of the sensors under different road scenarios.

[0108] Traditional systems employ fixed weights or linear weighting, failing to differentiate data priorities across different scenarios. For instance, in rainy or foggy weather, the confidence level of visual sensors decreases, yet traditional systems retain a high weight for these sensors, leading to false alarms. Therefore, this paper proposes a method to adjust the weight distribution of data in the fusion process based on environmental type. This process is a non-linear, dynamic allocation designed to adapt to environmental changes and improve the accuracy of the fusion process.

[0109] The original weights are the weights of each sensor used for multi-dimensional identification of vehicle road scenes. During data fusion processing, the sensor data and the corresponding original weights are usually combined to determine the final identification result of the current road scene, which is then used for subsequent blind spot warnings.

[0110] In some embodiments, adjusting the original weights of each sensor based on the environment type to obtain a first weight corresponding to each sensor includes: Based on the environment type, a preset exponent value for nonlinear operation is determined; the nonlinear operation is used to adjust the original weights to match the nonlinear characteristics of the sensor under different road scenarios. The sensor also includes a confidence level. Based on the index value and the confidence level of the sensor, a nonlinear calculation is performed to obtain a first weight corresponding to each of the sensors.

[0111] Specifically, index parameters are dynamically selected based on the environment type. k Adjust the weights. The nonlinear operation is represented by the following formula: , k=f (E) NV ) Among them, c i This represents the confidence value of the i-th data source, ranging from [0, 1], indicating the reliability of that data source in the current scenario. Example data source: High-precision map c map Visual perception c vision Vehicle status c vheicle , k This represents an environment-related index that is dynamically adjusted based on the scenario type, used to non-linearly amplify / reduce weight differences.

[0112] For confidence of all data sources k The sum of powers is used to normalize the weights.

[0113] From the above formula, it can be seen that when c i At higher levels, The growth rate is greater than linear (exponential). When c i At lower levels, The attenuation rate is amplified (exponential suppression). That is, the confidence level and the first weight are positively correlated.

[0114] By exponentially amplifying / reducing the weights of different data sources, the system adapts to the non-linear characteristics of the scenario and reduces the false alarm rate.

[0115] The following examples illustrate scenarios such as highways, construction sites, and urban areas. The specific implementation methods for other scenarios such as tunnels and nighttime are basically the same and will not be repeated here.

[0116] like Figure 5 As shown, Figure 5 This is a nonlinear weighting curve illustrated in this specification according to an exemplary embodiment.

[0117] If known, map: c map =0.9 (high confidence level in highway scenario).

[0118] Visual perception: c vision =0.8 (Rain and fog cause confidence to decrease).

[0119] V2X real-time data: c v2x =0.7 (Temporary data for the construction area, with moderate confidence).

[0120] Assuming a traditional linear assignment k=1.0, The map data weights are: w map =0.9 / (0.9+0.8+0.7)≈37.5% The weight of visual perception data is w vision =0.8 / (0.9+0.8+0.7)≈33.3% V2X data weighting is w v2x =0.7 / (0.9+0.8+0.7)≈29.2% This manual assigns a non-linear k=1.5, corresponding to the following highway segment scenario type: w map =0.91.5 / (0.91.5+0.81.5+0.71.5)≈39.6% w vision=0.81.5 / (0.91.5+0.81.5+0.71.5)≈33.2% w v2x =0.71.5 / (0.91.5+0.81.5+0.71.5)≈27.2% This manual assigns a non-linear k=1.2, corresponding to the following construction section scenario type: w map =0.91.2 / (0.91.2+0.81.2+0.71.2)≈38.3% w vision =0.81.2 / (0.91.2+0.81.2+0.71.2)≈33.3% w v2x =0.71.2 / (0.91.2+0.81.2+0.71.2)≈28.4% This manual assigns a non-linear k=0.8, corresponding to the urban road segment scenario type: w map =0.90.8 / (0.90.8+0.80.8+0.70.8)≈36.7% w vision =0.80.8 / (0.90.8+0.80.8+0.70.8)≈33.4% w v2x =0.70.8 / (0.90.8+0.80.8+0.70.8)≈29.9% The above examples demonstrate the importance of selecting the exponential parameter for different scenario types. k The changes in the weights of each sensor corresponding to the value.

[0121] As an example, the traditional linear assignment: k =1, high-speed scenario: k =1.5, increasing the weight of high-precision maps; Construction scenario: k =1.2, increasing the weight of V2X real-time data; Urban scenario: k =0.8, increasing visual weight; tunnel scene: k =1.3, increasing radar weight; Night scene: k =0.9, balancing visual and infrared data.

[0122] Real-time monitoring of sensor data quality, adjusting weight distribution using an environmentally relevant index k (e.g., in rain and fog scenarios). k =0.8, reducing visual weight and increasing LiDAR weight), dynamically reducing the impact of confidence data sources and lowering the false alarm rate of the system.

[0123] Based on training with historical data, the following is given:k The impact of value changes on the false alarm rate: (1) High-speed scenario k =1.5, estimated false alarm rate 4% In high-speed scenarios, c map =0.9, reliability is much higher than c vision =0.8, affected by rain and fog.

[0124] k =1.5 significantly improves map weight and suppresses the impact of visual false detections on decision-making.

[0125] The reason for the reduced false alarm rate is that visual sensors are prone to misdetecting obstacles in rain and fog (such as misidentifying raindrops as obstacles). Reducing their weight can reduce false triggers. Map data, on the other hand, is not affected by weather, and the exponential amplification of its weight makes it dominate the decision-making process, thus improving stability.

[0126] (2) Urban Scene k =0.8, estimated false alarm rate 7% Urban scenarios are complex (with dense pedestrian and vehicle traffic), requiring a balance between map, visual, and V2X data.

[0127] k =0.8 The advantage of compressing high-confidence data forces the system to rely on multi-sensor cross-validation.

[0128] It should be noted that, compared to high-speed scenarios k =1.5 for urban scenarios k The reason why the false alarm rate of 0.8 is higher than that in high-speed scenarios is that visual sensors are more critical in urban scenarios (such as pedestrian recognition), but are more susceptible to occlusion and lighting conditions. However, overall it is lower than that of traditional fixed sensors. k The false alarm rate is low.

[0129] Of course, if k Value too low k =0.5. Over-reliance on low-confidence data can lead to an increased false alarm rate. Therefore, in practical use, a trade-off needs to be struck based on actual requirements, and the setting should be adjusted to be more accurate. k Value. For k The optimization of values ​​will be explained in detail in later chapters.

[0130] (3) Traditional linear method with k=1.0, estimated false alarm rate 10%. Limitations: Linear weighting cannot distinguish scene characteristics; in rain and fog, the visual weight is still too high (e.g., 33.3%), leading to false alarms. Furthermore, the difference between map and visual weights is insufficient, easily triggering erroneous alarms when conflicts occur.

[0131] like Figure 6 As shown, Figure 6This is a comparative analysis diagram of the weight allocation in a construction scenario illustrated in this specification based on an exemplary embodiment.

[0132] As can be seen from the above examples, by adjusting... k Nonlinear dynamic allocation of sensor weights can effectively reduce the false alarm rate of the system.

[0133] Meanwhile, traditional systems require repeated conflict verification across multiple data sources, resulting in redundant decision-making processes. This solution, however, pre-determines the scenario type, narrowing the decision-making scope. Subsequently, through training with historical data, it can quickly match the optimal weight coefficients, reducing real-time computation time and improving the system's execution and control response speed.

[0134] Therefore, the environment type of the road scene is first determined, and the weights are dynamically adjusted according to the environment type to obtain the first weight. By amplifying or reducing the weights of different data sources, the nonlinear characteristics of the scene can be adapted.

[0135] Based on the above embodiments, the system can adapt to environmental changes in real time through the dynamic weight adjustment mechanism of each data source, maximizing the collaborative benefits of multimodal data while ensuring functional safety.

[0136] In step 106, the sensor data is fused according to the first weight, and blind spot warning is performed based on the fusion result.

[0137] The sensor data includes initial recognition results obtained after multi-dimensional identification of the vehicle road scene. After obtaining the weights of each data source, the data is fused based on the initial recognition results and the first weight corresponding to each data source to determine the legality of the road scene within the vehicle's blind spot, i.e., to indicate whether the current scene belongs to the fusion result of a non-changeable lane scene.

[0138] Specifically, the legality of road scenarios within the vehicle's blind spot is determined by the degree of support for scenarios where lane changes are not allowed in the current scenario.

[0139] As an example, the result of identifying a non-reversible lane scenario is denoted as the first value, and the result of identifying a reversible lane scenario is denoted as the second value. By summing the identification results and weights from each data source, the degree of support for the conclusion that the current road scenario belongs to a non-reversible lane scenario is obtained.

[0140] The probability of a road scene belonging to a non-changeable lane scene is divided into three levels: high, medium, and low. Different levels correspond to different numerical ranges. By matching the comprehensive calculation results with each numerical range, the road scene recognition result is obtained.

[0141] For example, if the probability P that cannot be changed lanes cclIs high, indicating a high degree of certainty that the road scenario is a non-lane-changing scenario, emphasizing an absolute prohibition of lane changes. Subsequently, based on the mapping relationship between the hierarchical warning control strategy and the scenario in the pre-decision layer, the corresponding suppression warning strategy is determined.

[0142] As an example, the non-lane-changing probability P ccl One case of the initial recognition results of each data source corresponding to being high is: ① The high-precision map is marked as a non-lane-changing area (ramp / construction area, etc.); ② High-confidence visual verification: The lane line type is a solid line; ③ V2X does not push a lane change permission instruction; The non-lane-changing probability P ccl Is medium, indicating that the current road scenario is likely to be a non-lane-changing scenario with uncertainty. For example, in the case of a tidal lane, etc., dynamic response is required to determine whether lane changes are allowed. Subsequently, based on the mapping relationship between the hierarchical warning control strategy and the scenario in the pre-decision layer, the corresponding limited warning strategy is determined to narrow the monitoring range and reduce the warning frequency.

[0143] As an example, the non-lane-changing probability P ccl One case of the initial recognition results of each data source corresponding to being medium is: ① The map is not marked as non-lane-changing; ② Visual detection shows alternately solid and dashed lane lines or a tidal lane sign; ③ The V2X dynamic traffic sign allows lane changes; The non-lane-changing probability P ccl Is low, indicating that it is certain that the road scenario is a lane-changing scenario or it is impossible to determine a prohibition of lane changes, and the default lane-changing state is allowed. Subsequently, based on the mapping relationship between the hierarchical warning control strategy and the scenario in the pre-decision layer, the corresponding normal warning strategy is determined to execute the original blind spot warning function.

[0144] As an example, the non-lane-changing probability P ccl One case of the initial recognition results of each data source corresponding to being low is: ① The map is marked as a lane-changing area; ② Visual detection shows a dashed lane line; ③ There is no V2X prohibition instruction.

[0145] Through the above embodiments, this multi-dimensional data-driven scenario legality recognition mechanism breaks through the limitations of traditional single-sensor judgment. Through the prior knowledge constraint of the high-precision map, the real-time verification of visual perception, and the dynamic association of vehicle states, the full process from data collection to decision-making output can be completed in a short time, ensuring that in complex scenarios such as ramp merging and solid-line lanes, the blind spot warning system can accurately distinguish legal lane changes from illegal operations, laying a data foundation for the dynamic adjustment of subsequent warning strategies.

[0146] In some embodiments, the step of fusing the sensor data according to the first weight and performing blind spot warning based on the fusion result includes: The sensor data is fused according to the first weight to obtain a fusion result indicating the road scene within the blind spot of the vehicle; Based on the fusion results, an early warning control strategy is selected from multiple preset hierarchical early warning control strategies, and the working state of the blind spot early warning function is dynamically scheduled according to the selected early warning control strategy. When the blind spot early warning function is activated, risk monitoring is performed continuously.

[0147] The warning control strategy described in this article refers to the behavioral scheduling of the blind spot warning function configured in the vehicle, including but not limited to warning output, parameter configuration, and actuator linkage. Even when the warning control strategy is executed, the vehicle's blind spot warning function continues to monitor blind spots and calculate risks. For example, when the blind spot warning function is scheduled to operate using a suppression warning strategy, only the active warning output to the driver is turned off, while maintaining background risk monitoring capabilities.

[0148] As an example, the early warning control strategy includes a suppression early warning strategy, a limited early warning strategy, and a normal early warning strategy.

[0149] The suppression and early warning strategy includes temporarily suppressing blind spot warnings, and the HMI displays a "No Lane Change" icon.

[0150] For example, traditional blind spot warning functions have poor scenario adaptability, continuing to issue warnings in road structures where lane changes are not allowed or in solid-line lane scenarios, leading to driver misoperation or interference. However, by using a warning suppression strategy to schedule the behavior of the blind spot warning function, the output of the blind spot warning is temporarily suppressed, and a "no lane change" icon is displayed through the HMI. In this way, the driver will not be disturbed by the warning in scenarios where lane changes are not allowed (such as solid-line areas), while the vehicle's continuous back-end warning monitoring also ensures vehicle safety.

[0151] It should be noted that when the hands-free driving warning system and the driver attention monitoring system detect that the driver is in an abnormal driving state, the suppression warning strategy should be prohibited or the suppression warning should be canceled. An abnormal driving state includes, but is not limited to, the driver taking their hands off the wheel or having their gaze deviate from the road ahead.

[0152] The limited early warning strategy is based on the original early warning interaction logic, which cancels the audio early warning and increases the early warning threshold.

[0153] For example, the visual warning delay can be increased to 1 second (to filter false warnings of less than 1 second), and the duration of the alarm for a single target can not exceed 3 seconds, in order to minimize interference with the driver and reduce the false alarm rate.

[0154] The normal warning strategy is based on the TTC calculation of the target and the vehicle to determine whether the blind spot warning function is triggered. When the blind spot warning function is activated, it responds according to the original function.

[0155] In summary, the blind spot early warning function, after being scheduled through a tiered early warning control strategy, operates in the following order from high to low intensity: suppression early warning, limited early warning, and normal early warning.

[0156] Based on the aforementioned scene recognition results, an adaptive scene warning and control strategy is selected. Specifically, a suppression warning strategy is selected for scenes where lane changes are not possible, a normal warning strategy is selected for scenes where lane changes are possible, and a limited warning strategy is selected for scenes where the determination of lane changes is uncertain.

[0157] Through the above embodiments, the blind spot warning function is scheduled by a hierarchical warning control strategy. The blind spot warning function is controlled to adaptively adjust according to the road scenario. In scenarios where lane changes are not allowed, the warning is suppressed to avoid false alarms.

[0158] In some embodiments, after selecting an early warning control strategy from a plurality of preset hierarchical early warning control strategies based on the fusion result, and dynamically scheduling the working state of the blind spot early warning function according to the selected early warning control strategy, the method further includes: Monitor vehicle status parameters and determine whether the driver intends to change lanes in the current road scenario based on the monitored data; If a lane-changing intention is determined, the execution intensity of the blind spot warning function is increased after dynamic scheduling with the selected warning control strategy.

[0159] To improve a vehicle's obstacle avoidance capabilities, proactive intervention is conducted when the driver intends to change lanes, thereby enhancing vehicle safety.

[0160] As an example, the driver's lane-changing intention is determined based on the driver's actions. These actions include, but are not limited to, activating the turn signal and turning the steering wheel; this information is collected by vehicle sensors.

[0161] When it is determined that there is an intention to change lanes, the vehicle should be judged whether to respond to the lane change operation corresponding to the intention to change lanes based on the real-time parameters of the vehicle. Based on the judgment result, an execution strategy is selected from multiple preset execution strategies, and the execution intensity of the blind spot warning function is increased by the selected execution strategy.

[0162] The execution strategy includes enhanced early warning strategy and proactive intervention strategy. By further judging the lane change operation corresponding to the lane change intention, different lane change intentions are identified, and the corresponding execution strategy is selected according to the degree of the lane change intention.

[0163] A driver's intention to change lanes can be categorized into three situations.

[0164] The first scenario involves the driver's intention to change lanes. The criteria for this are: the driver only uses their turn signal but has no actual intention to forcibly change lanes.

[0165] If the result of the judgment is negative, the enhanced early warning strategy is selected to increase the execution intensity of the blind spot early warning function, and the execution intensity includes the early warning output intensity.

[0166] For example, the existing alarm prompts can be enhanced, including but not limited to doubling the frequency of the alarm sound, providing more prominent visual prompts, and adding additional tactile prompts.

[0167] The second scenario involves a driver's intent to forcibly change lanes. The criteria for this are: the steering wheel angle or the vehicle's real-time yaw rate exceeds a set value. For example, the absolute value of the steering wheel angle is greater than 10° and the vehicle's real-time yaw rate is greater than 5° per second for 3 seconds.

[0168] If the determination is yes, the active intervention strategy is selected, and the vehicle control system generates an operation command opposite to the lane-changing operation to assist the vehicle's movement. The control includes combined lateral and longitudinal control.

[0169] For example, lateral control: a counter-torque is generated by the EPS system to counteract the driver's steering input. The torque increases linearly with the steering wheel angle. For instance, if the steer-by-wire system applies a counter-torque (10 N·m), the brake-by-wire system brakes one wheel (deceleration of 0.3g).

[0170] However, if the driver continuously applies steering force or a sudden, significant steering torque jerk (measuring the abruptness of steering input) occurs, the driver overriding control logic is executed, and the vehicle system disengages lateral control intervention to balance the safety of intelligent control and manual driving. Further details will not be elaborated here.

[0171] By monitoring the jerk value, the system can achieve more flexible human-machine collaboration logic and avoid intervention conflicts when the driver has a strong intention to operate.

[0172] Longitudinal control: Prioritizes actively increasing braking pressure to reduce braking clearance and shorten driver reaction time. Furthermore, at lower speeds, such as less than or equal to 60 km / h, the vehicle stability control system brakes the outer wheels to generate yaw moment and correct the vehicle's position. At higher speeds, such as greater than 70 km / h, all-wheel braking is applied to avoid emergency collisions.

[0173] In the third scenario, if the driver has no intention of changing lanes or forcibly changing lanes, then no enhanced warning or proactive intervention will be issued.

[0174] It should be noted that the execution of enhancement and intervention strategies depends on the vehicle state / vehicle state parameters, and the confidence level of these parameters must be guaranteed. If the confidence level of the vehicle state parameters is significantly abnormal (the confidence level C of the vehicle state data...), then... vehicle If the value is ≤0.3, then the enhancement and intervention strategies are prohibited. Furthermore, the early warning control strategy will be forcibly downgraded to a limited early warning mode.

[0175] Through the above embodiments, the hierarchical early warning control strategy and the linkage design of the actuator reduce the risk of vehicle lane changes, reduce false alarms, and achieve a balance between safety (preventing dangerous lane changes) and driver experience (reducing unnecessary interference).

[0176] Through the above embodiments, the legality of road scenes within the vehicle's blind spot range is identified in real time based on multi-dimensional data to determine whether the current scene belongs to the fusion result of a non-changeable lane scenario. Based on the identified scene, a warning control strategy is selected from multiple pre-set hierarchical warning control strategies. The selected warning control strategy dynamically schedules the current blind spot warning function's working state, achieving scene adaptation of the current blind spot warning function, reducing the probability of false alarms, and providing accurate warnings when needed, thereby balancing vehicle safety and driver experience.

[0177] In some embodiments, after fusing the sensor data according to the first weight and performing blind spot warning based on the fusion result, the method further includes: Obtain the results of early warning and intervention; The exponential value of the preset nonlinear operation is optimized based on the early warning intervention result; the nonlinear operation is used to adjust the original weights to match the nonlinear characteristics of the sensor under different road scenarios.

[0178] During a vehicle driving cycle, and when driving under the same environmental type, the original weights of each sensor are adjusted based on the optimized index value to obtain a first weight corresponding to each sensor.

[0179] After issuing a blind spot warning, a closed-loop feedback optimization mechanism is triggered.

[0180] Adjustments are made based on historical early warning intervention results using reinforcement learning modules. k The values ​​are adjusted to optimize the weight parameters of each sensor, enabling the weight parameters to evolve automatically, thereby reducing the cost of manual parameter tuning and improving system adaptability.

[0181] As an example, historical early warning intervention results include success rate, false alarm rate, and energy efficiency. These results can be obtained by monitoring vehicle status during or after the implementation of early warning control strategies.

[0182] Reward function: R = γ1 * S 成功 −γ2*F 误报 +γ3*E 能耗 Among them, S 成功 : Intervention success rate [0, 1], which is the percentage of active collision avoidance or correct suppression of false alarms.

[0183] F 误报 False alarm rate [0, 1]: The proportion of alarms triggered erroneously when there is no risk.

[0184] E 能耗 Energy efficiency [0, 1], computing system resource utilization rate (e.g., the lower the CPU utilization, the higher the score).

[0185] γ1: Success rate weight, emphasizing the priority of security; γ2: False alarm rate penalty term, suppressing oversensitivity; γ3: Energy efficiency incentive item to encourage resource optimization.

[0186] Maximizing rewards: Adjusting the exponential value of nonlinear operations through reinforcement learning agents. k This allows the system to strike a balance between security (high success rate), user experience (low false alarms), and resource efficiency (low energy consumption).

[0187] For example, if S makes a decision in a certain instance... 成功 =1,F 误报 =0, E 能耗 =0.8, γ1=0.7, γ2=-0.3, γ3=0.1. Therefore, the reward function R=0.7*1-0.3*0+ 0.1*0.8=0.78, which drives... k Evolve towards the optimal value.

[0188] As an example, when the reward function R ≥ a set value, it is permissible to increase the transfer ratio α and adjust the index value. k Among them, adjusting environmentally related index values. k ( k±0.1), adjusting the weight transfer coefficient α within the range [0.2, 0.4]. For example, when the reward function R ≥ the set value, k Increase by 0.1 when R < the set value. k Reduce by 0.1, and set the maximum adjustment amount Δ according to actual needs. k .

[0189] Similarly, the reinforcement learning module dynamically adjusts the transfer ratio α based on historical intervention results to obtain early warning intervention results; the early warning results include the early warning success rate, false alarm rate, and energy efficiency after executing blind zone early warning; the set transfer ratio is optimized based on the early warning intervention results.

[0190] For the adjusted index value k Compared to the transfer ratio α, it is more in line with the current driving state of the vehicle and the accuracy of the blind spot warning system. Therefore, within the set period, when it is determined that the environmental type of the vehicle road scene is the same as the environmental type of the optimized parameters, the original weights of each sensor are directly adjusted using the optimized parameters to obtain the first weight corresponding to each sensor. The sensor data is then fused based on the first weight, and blind spot warning is performed based on the fusion result.

[0191] The set period can be, but is not limited to, the vehicle driving cycle, the cycle from vehicle power-on to power-off, or a user-defined time cycle.

[0192] This specification provides a blind spot warning method, device, and computer-readable storage medium. It performs multi-dimensional identification of vehicle road scenes, obtaining data from multiple sensors to determine the environmental type of the road scene where the vehicle is located. Based on the environmental type, the original weights of each sensor are adjusted to obtain a first weight corresponding to each sensor. By identifying the environmental type, weights are dynamically allocated non-linearly based on an environment-related index to adapt to complex scenarios. Sensor data is fused according to the first weight, dynamically reducing the impact of low-confidence data sources on the fusion process. Blind spot warning is performed based on the fusion result, effectively reducing the false alarm rate. Furthermore, the dynamic weight adjustment based on the environment-related index improves the response speed of the system's execution control.

[0193] like Figure 7 As shown, Figure 7 This is a block diagram illustrating an apparatus according to an exemplary embodiment, the apparatus comprising: The data acquisition module is used to perform multi-dimensional identification of vehicle road scenes and obtain data from multiple sensors; The environment classification module is used to determine the environment type of the road scene where the vehicle is located; A nonlinear weight allocation module is used to adjust the original weights of each sensor based on the environment type to obtain a first weight corresponding to each sensor. The blind spot warning module is used to perform fusion processing on the sensor data according to the first weight, and to perform blind spot warning based on the fusion result.

[0194] The specific implementation process of the functions and roles of each module / submodule / unit in the above device can be found in the implementation process of the corresponding steps in the above method, which can achieve the same technical effect, and will not be repeated here.

[0195] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of the solution in this specification according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0196] Figure 8 An example is a schematic diagram of the physical structure of a blind spot early warning device, such as... Figure 8 As shown, the blind spot warning device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, communication interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions from the memory 830 to execute the blind spot warning method.

[0197] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 described in the various embodiments of this application. 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.

[0198] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the blind spot warning method provided by the above methods.

[0199] In another aspect, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the blind zone warning methods provided by the above methods.

[0200] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0201] Other embodiments of this specification will readily occur to those skilled in the art upon consideration of the specification and practice of the invention claimed herein. This specification is intended to cover any variations, uses, or adaptations that follow the general principles of this specification and include common knowledge or customary techniques in the art not claimed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this specification are indicated by the following claims.

[0202] It should be understood that this specification 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. The scope of this specification is limited only by the appended claims.

[0203] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A blind area warning method characterized by comprising: The method comprises: Multi-dimensional recognition is performed on a vehicle road scene to obtain a plurality of sensor data; Original weights of each sensor are adjusted based on an environment type to obtain first weights corresponding to each sensor; The sensor data is fused according to the first weights, and blind area warning is performed based on a fusion result.

2. The blind zone warning method of claim 1, wherein, The original weights of each sensor are adjusted based on the environment type to obtain the first weights corresponding to each sensor, which comprises: According to the environment type, an index value of a preset nonlinear operation is determined; the nonlinear operation is used to adjust the original weights to match the nonlinear characteristics of the sensors in different road scenes; The sensor also includes a confidence level, and the first weights corresponding to each sensor are obtained by performing nonlinear operation based on the index value and the confidence level of the sensor.

3. The blind zone warning method of claim 2, wherein, In the nonlinear operation, the confidence level is positively correlated with the first weights.

4. The blind zone warning method of claim 1, wherein, Before the original weights of each sensor are adjusted based on the environment type, the method further comprises: Environment data related to environment recognition is obtained from the sensor data; An initial environment type of a road scene where a vehicle is located is identified according to the environment data and a corresponding second weight, to obtain an initial environment type; When the confidence level of the initial environment type is greater than or equal to a confidence threshold, the initial environment type is determined as the final environment type of the road scene where the vehicle is located.

5. The blind zone warning method of claim 4, wherein, The method further comprises: When the confidence level of the initial environment type is less than a set threshold, a sensor whose confidence level is less than an abnormal threshold is determined as an abnormal sensor; A second weight of the abnormal sensor is distributed to a backup sensor according to a set transfer ratio, to determine a third weight of the abnormal sensor after distribution and a fourth weight of the backup sensor after distribution; The environment type of the road scene where the vehicle is located is identified based on the environment data and the third weight of the abnormal sensor, the environment data and the fourth weight of the backup sensor, and the environment data and the second weight of other sensors, to obtain the final environment type of the road scene where the vehicle is located.

6. The blind zone warning method of claim 5, wherein, The method further comprises: When the confidence level of the abnormal sensor meets a weight recovery condition, the third weight of the abnormal sensor is updated to the second weight of the abnormal sensor before distribution; The weight recovery condition comprises that the confidence level of the abnormal sensor is continuously greater than the abnormal threshold within a set time.

7. The blind spot warning method according to any one of claims 4 to 6, characterized by, The confidence level of the initial environment type is obtained in the following manner: The confidence levels of each of the environment data are weighted and fused to determine the confidence level of the initial environment type.

8. The blind zone warning method of claim 1, wherein, The sensor data is fused according to the first weights to obtain a fusion result indicating a road scene in a blind area of the vehicle; and blind area warning is performed based on the fusion result. ​ According to the fusion result, an early warning control strategy is selected from a plurality of preset hierarchical early warning control strategies, and a working state of the blind area early warning function is dynamically scheduled by using the selected early warning control strategy, wherein the blind area early warning function is activated to continuously perform risk monitoring.

9. The blind zone warning method of claim 1, wherein, After the fusion processing of the sensor data according to the first weight and the blind area early warning based on the fusion result, the method further comprises: obtaining an early warning intervention result; optimizing an index value of a preset nonlinear operation based on the early warning intervention result, wherein the nonlinear operation is used to adjust an original weight to match a nonlinear characteristic of the sensor in different road scenes; in a vehicle driving cycle and when driving in the same environment type, adjusting the original weight of each sensor based on the optimized index value to obtain a first weight corresponding to each sensor.

10. A computer device, comprising: The computer readable storage medium stores a blind area early warning program, and the blind area early warning program implements the steps of the blind area early warning method when executed.

11. A computer readable storage medium, characterized in that, The computer readable storage medium stores a blind area early warning program, and the blind area early warning program implements the steps of the blind area early warning method when executed.

Images