Interventional control method for aided driving and automobile

Abstract

The invention provides an intervention control method for aided driving and an automobile, and the method comprises the steps: obtaining a panoramic stitching image of a test environment, and recognizing a calibration object set in the panoramic stitching image; generating a semantic object according to the calibration object set; obtaining pose data of the vehicle, and determining a spatial relationship parameter between the vehicle and the semantic object according to the semantic object and the pose data; performing intention classification according to the semantic object and the driving state information of the user to obtain a driving intention identifier; and determining whether the vehicle is in danger according to the semantic object, the space calibration data and the driving intention identifier, and intervening in controlling the vehicle when the vehicle is in danger.

Description

Technical Field

[0001] The present application relates to the field of assisted driving technology, and in particular to an intervention control method and a vehicle for assisted driving. Background Art

[0002] Traditional intervention control methods rely on a large amount of physical calibration equipment and complex site layouts during the testing process, making it difficult to flexibly construct diverse test scenarios. In addition, existing methods have limited capabilities in driving intention recognition and find it difficult to combine the driver's true intentions with environmental information for comprehensive decision-making. This can easily lead to misjudgments or inappropriate intervention timing, seriously affecting the effectiveness and reliability of assisted driving system testing. Summary of the Invention

[0003] The present application provides an intervention control method and a vehicle for assisted driving, which are used to improve the effectiveness and reliability of assisted driving system testing.

[0004] In a first aspect, an embodiment of the present application provides an intervention control method for assisted driving, the method comprising: Acquire a panoramic stitched image of the test environment, and identify a set of calibration objects in the panoramic stitched image; generating a semantic object according to the set of calibration objects; Acquiring posture data of a vehicle, and determining spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the posture data; performing intention classification based on the semantic object and the user's driving state information to obtain a driving intention identifier; Determine whether the vehicle is in danger based on the semantic object, the spatial calibration data and the driving intention identifier, and intervene to control the vehicle when danger occurs.

[0005] In a second aspect, an embodiment of the present application provides a car, characterized in that the car is used to execute the intervention control method for assisted driving as described in any one of the embodiments of the present application.

[0006] An embodiment of the present application provides an intervention control method for assisted driving, the method comprising: obtaining a panoramic stitched image of a test environment, identifying a set of calibration objects in the panoramic stitched image; generating a semantic object based on the set of calibration objects; obtaining vehicle posture data, determining spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the posture data; performing intention classification based on the semantic object and the user's driving status information to obtain a driving intention identifier; determining whether the vehicle is in danger based on the semantic object, the spatial calibration data, and the driving intention identifier, and intervening to control the vehicle when danger occurs. In the above method, by obtaining a panoramic stitched image of the test environment and identifying a set of calibration objects therein to generate a semantic object, it is achieved that calibration objects are constructed through images, spatial relationship parameters with semantic objects are determined by combining vehicle posture data, and driving intention identifiers are obtained by performing intention classification based on the user's driving status information. The comprehensive risk judgment mechanism based on semantic objects, spatial calibration data, and driving intention identifiers avoids the problems of misjudgment or inappropriate intervention timing, and significantly improves the effectiveness and reliability of intervention control of the assisted driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of an intervention control method for assisted driving provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0010] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0011] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0012] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0013] See also Figure 1 , Figure 1 This is a schematic flow chart of an intervention control method for assisted driving provided by an embodiment of the present application. Figure 1 As shown, the specific steps of the intervention control method for assisted driving include: S101-S105.

[0014] S101: Obtain a panoramic stitched image of a test environment, and identify a set of calibration objects in the panoramic stitched image.

[0015] For example, multiple high-resolution cameras are used to capture images of the test environment from different angles. During the acquisition process, the camera positions are kept fixed and the coverage areas are fully overlapped. The collected multi-view image data are geometrically aligned using a feature point matching algorithm, and pixel-level alignment is achieved by calculating the transformation matrix between the images. After the alignment is completed, the overlapping areas are fused to eliminate seams and brightness differences, and a panoramic stitching image covering the entire test environment is generated. Edge detection and shape recognition algorithms are used in the panoramic stitching image to locate candidate areas for calibration objects, and contour extraction and geometric feature analysis are performed on the candidate areas. According to the predefined geometric features of the calibration objects, including the checkerboard pattern of the square calibration plate, the circular contour of the circular markers, and the linear features of the linear calibration ruler, the identified objects are classified and verified. By calculating the center coordinates, size parameters, and direction angles of the calibration objects, a unique identification code is established for each calibration object, and the information of all verified calibration objects is integrated to form a calibration object set.

[0016] S102: Generate semantic objects according to the calibration object set.

[0017] Based on the spatial coordinates and geometric parameters of each calibration object in the calibration object set, a three-dimensional spatial coordinate system for the test environment is established, with the coordinate origin and axis determined by the distribution pattern of the calibration objects. The definition rules for virtual test objects are read according to a preset test configuration file, which contains the generation conditions and attribute parameters for static obstacles, dynamic targets, and path guidance points. A spatial mapping algorithm is used to convert the two-dimensional image coordinates of the calibration objects into three-dimensional spatial coordinates, and the corresponding virtual test objects are generated at the specified locations. Attributes are assigned to the generated virtual test objects: fixed positions and collision boundaries are set for static obstacles, motion trajectories and speed parameters are configured for dynamic targets, and navigation priorities are assigned to path guidance points. A semantic annotation algorithm is used to assign a specific object type label, motion state description, and interaction priority value to each virtual test object. The geometric information, motion parameters, and semantic labels of the virtual test objects are data-encapsulated to form semantic objects containing complete attribute information.

[0018] S103 : Acquire the position and posture data of the vehicle, and determine the spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the position and posture data.

[0019] Onboard sensors collect the vehicle's position coordinates, heading, pitch, and roll angles in real time within the test environment. The vehicle's linear and angular velocity data are also acquired to form complete pose data. Positional state data, including the vehicle's 3D coordinates, boundary dimensions, shape outline, and motion state, is extracted from semantic objects. Based on the vehicle's pose data and the semantic object's positional state data, vector calculations are used to determine the Euclidean distance, relative azimuth, and approach velocity between the vehicle and each semantic object. Taking into account the static attributes of the semantic objects, the spatial geometric constraints of the vehicle relative to each object are calculated, including minimum safe distance, traversable area, and potential collision path. Combining dynamic and static spatial relationship parameters, comprehensive spatial relationship parameters are generated to describe the relative position, motion trends, and interaction potential between the vehicle and all semantic objects. These spatial relationship parameters provide an accurate spatial geometric basis for subsequent hazard assessment and intervention decisions.

[0020] S104: performing intention classification based on the semantic object and the user's driving state information to obtain a driving intention identifier.

[0021] The system collects multi-dimensional user operational data during driving, including steering wheel angle and its rate of change, accelerator and brake pedal position information, gear shift sequence, and gaze focus distribution captured by eye tracking. Temporal features are extracted from this collected driving state information to analyze the frequency, amplitude, and duration patterns of operational actions. Based on the spatial distribution characteristics and interaction priorities of semantic objects, a correlation weight matrix is ​​constructed to reflect the influence of different objects on driving decisions. By analyzing the correspondence between the user's gaze focus and the position of semantic objects, the user's attention distribution for each semantic object is calculated, identifying the primary interactive objects in the current driving scenario. The temporal features are fused with the correlation weight matrix, and the user's attention distribution and operational behavior patterns are combined to infer the user's driving strategy intention. A multimodal feature fusion algorithm is used to generate a hierarchical driving intention signature comprising operational, strategic, and target levels. Each intention level is assigned a corresponding confidence score.

[0022] S105: Determine whether the vehicle is in danger based on the semantic object, the spatial calibration data, and the driving intention identifier, and intervene to control the vehicle when danger occurs.

[0023] For example, a dynamic safety envelope is established for each semantic object based on its geometric and motion characteristics. This safety envelope adjusts in real time based on the object's motion state and vehicle proximity. Distance, angle, and velocity information in the spatial relationship parameters are used to calculate the collision time, collision probability, and potential collision severity between the vehicle and each semantic object, generating a multi-dimensional risk assessment. The probability distribution information in the driving intention identifier is combined to establish a mapping between intention and risk. Different driving intentions are assigned different risk weights, dynamically adjusting the risk assessment threshold. A sliding window approach is used to analyze the risk assessment values ​​over time to predict risk trends and potential dangerous moments. When the comprehensive risk index exceeds a preset threshold, the vehicle is deemed to be in collision danger. The severity of the risk determines the level of intervention control, including warnings, assisted intervention, and forced takeover. Based on the real-time risk assessment results and vehicle dynamic constraints, a smooth intervention control trajectory is generated. Intervention control is implemented through braking, steering, or speed adjustment, and the effectiveness of the control is continuously monitored until the dangerous state is resolved.

[0024] An embodiment of the present application provides an intervention control method for assisted driving, the method comprising: obtaining a panoramic stitched image of a test environment, identifying a set of calibration objects in the panoramic stitched image; generating a semantic object based on the set of calibration objects; obtaining vehicle posture data, determining spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the posture data; performing intention classification based on the semantic object and the user's driving status information to obtain a driving intention identifier; determining whether the vehicle is in danger based on the semantic object, the spatial calibration data, and the driving intention identifier, and intervening to control the vehicle when danger occurs. In the above method, by obtaining a panoramic stitched image of the test environment and identifying a set of calibration objects therein to generate a semantic object, it is achieved that calibration objects are constructed through images, spatial relationship parameters with semantic objects are determined by combining vehicle posture data, and driving intention identifiers are obtained by performing intention classification based on the user's driving status information. The comprehensive risk judgment mechanism based on semantic objects, spatial calibration data, and driving intention identifiers avoids the problems of misjudgment or inappropriate intervention timing, and significantly improves the effectiveness and reliability of intervention control of the assisted driving system.

[0025] In some embodiments, obtaining a panoramic stitched image of the test environment and identifying a set of calibration objects in the panoramic stitched image include: shooting multi-perspective image data of the test environment; aligning the multi-perspective image data to generate a panoramic stitched image of the test environment; identifying calibration objects in the panoramic stitched image, the calibration objects including: a grid calibration plate, circular marking points, and a straight calibration ruler; detecting the calibration objects in the panoramic stitched image to form a set of calibration objects.

[0026] For example, multiple high-resolution cameras installed at different locations in the test environment are used to take synchronous photos, ensuring that there is an overlap of more than 30% between the fields of view of adjacent cameras, thereby obtaining multi-view image data covering the entire test environment. In the image registration process, the SIFT feature point detection algorithm is used to find corresponding feature point pairs between adjacent images, and the rotation matrix and translation vector between the images are determined by calculating the geometric transformation relationship between the feature points. Based on the feature point matching results, the multi-view image data is geometrically corrected and perspective transformed, and the images of different perspectives are unified into the same coordinate system. The brightness and color differences in the overlapping areas are processed by multi-band fusion technology. The registered images are seamlessly stitched to generate a high-resolution panoramic stitching image, which fully displays the spatial layout of the test environment and the distribution of calibration objects. The edge detection algorithm is applied to the panoramic stitching image to extract the boundary contours of potential calibration objects, the Hough transform is used to detect straight and circular features, and the corner detection algorithm is used to identify the grid pattern. Grid calibration plates are identified by their unique black and white checkerboard pattern, and their validity is verified by calculating the grid point spacing and arrangement pattern. Circular markers are selected by analyzing the circularity and area ratio of their outlines, and linear calibration rulers are confirmed by detecting continuous straight line segments and scale marks. Each identified calibration object undergoes precise contour extraction and geometric parameter calculation, including center coordinates, size, orientation angle, and shape characteristics. Each calibration object is assigned a unique number and type label, and the information of all verified calibration objects is aggregated into a calibration object set containing position, geometry, and type information.

[0027] In some embodiments, generating semantic objects based on the calibration object set includes: establishing a three-dimensional spatial coordinate system and virtual test objects of the test environment based on the calibration object set and a preset test configuration file, wherein the virtual test objects include static obstacles, dynamic targets, and path guide points; performing semantic classification and annotation on the virtual test objects to determine attribute information of each virtual test object, wherein the attribute information includes object type, motion state, and interaction priority; and generating semantic objects based on the virtual test objects and the attribute information.

[0028] For example, the spatial distribution characteristics of each calibration object in the calibration object set are used to select an appropriate coordinate origin and axial direction. The spatial arrangement pattern of the calibration objects is fitted using the least squares method to establish a unified three-dimensional coordinate system. A preset test configuration file stores the rules for generating virtual test objects in XML format, including detailed definitions of the number distribution, position constraints, size ranges, and behavioral parameters of each object type. Virtual test objects are generated within a designated area within the three-dimensional coordinate system according to the rules in the configuration file. Static obstacles are placed in fixed locations to simulate immovable objects such as road guardrails and traffic signs. Dynamic targets are set along the motion path to simulate the behavior of other vehicles or pedestrians. Path guidance points are distributed along a predetermined trajectory to guide the vehicle's desired route. A semantic annotation program assigns each virtual test object a specific object type identifier, including information such as vehicle, pedestrian, guardrail, and signboard. The object's motion state is also determined based on its preset behavioral pattern. Static objects are marked as stationary, while dynamic objects are marked as uniform, accelerating, decelerating, or reciprocating based on their motion trajectory. Based on the safety requirements and traffic regulations of the test scenario, interaction priorities are assigned to each virtual test object, with emergency avoidance objects receiving the highest priority, regular traffic participants receiving a medium priority, and background environmental objects receiving a lower priority. The virtual test object's geometric information, spatial coordinates, and motion parameters are then encapsulated and bound to its attributes, including object type, motion state, and interaction priority, to form semantic objects containing complete descriptive information. These semantic objects provide structured environmental information for the vehicle's perception decisions.

[0029] In some embodiments, the acquiring of the vehicle's posture data and the determining of the spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the posture data include: extracting the position state data and static attribute information of the semantic object, the static attribute information including: object size, shape outline and fixed position coordinates; calculating the dynamic spatial relationship between the vehicle and the semantic object based on the posture data and the position state data, the dynamic spatial relationship parameters including: relative distance, relative angle and relative speed; calculating the spatial constraint relationship of the vehicle relative to the semantic object based on the static attribute information to acquire the static spatial relationship; generating spatial relationship parameters based on the dynamic spatial relationship parameters and the static spatial relationship.

[0030] For example, a high-precision integrated navigation device acquires real-time six-degree-of-freedom (6DOF) pose data within the test environment, including 3D position coordinates, heading, pitch, and roll angles, while also recording changes in the vehicle's linear and angular velocity. From the semantic object data structure, the system extracts each object's real-time position coordinates, velocity, acceleration, and other positional state data. Static attribute information, such as the object's length, width, and height dimensions, boundary outline, and collision detection boundary, is also obtained. Vector calculation methods are used to calculate the Euclidean distance between the vehicle's center of mass and the geometric center of each semantic object. The vehicle's heading angle and the object's relative position are used to calculate the relative azimuth and pitch angles in the vehicle coordinate system. The relative approach and departure speeds are calculated based on the velocity vector difference between the vehicle and the object. The closest and farthest distances between the vehicle and the object's boundary are calculated based on the semantic object's shape and outline information. The minimum clearance required for safe passage is determined based on the object's dimensions and vehicle geometry. Based on fixed position coordinates, the vehicle's passable area and potential conflict zone relative to the static object are analyzed. The relative distance, relative angle, and relative speed in the dynamic spatial relationship parameters are comprehensively processed with the geometric constraints, safety gaps, and traffic area information in the static spatial relationship to generate spatial relationship parameters that describe the spatial geometric relationship, motion trend relationship, and interaction possibility between the vehicle and all semantic objects. These parameters provide an accurate spatial geometric basis for subsequent behavior prediction and risk assessment.

[0031] In some embodiments, the intention classification is performed based on the semantic object and the user's driving status information to obtain a driving intention identification, including: selecting a target semantic object that has an interactive relationship with the vehicle from the semantic object; identifying the user's possible behavior types based on the driving status information, and the possible behavior types include: going straight, changing lanes, turning, parking and reversing; calculating a probability distribution based on the target semantic object and the possible behavior types; determining a target behavior type from the possible behavior types based on the maximum probability value in the probability distribution; and determining a driving intention identification based on the target behavior type, the driving intention identification.

[0032] For example, by analyzing the spatial distance relationship and relative motion trends between semantic objects and the vehicle, target semantic objects with potential for interaction are identified within a preset range from the vehicle. These objects are typically located in the vehicle's forward path or in adjacent lanes and have a high interaction priority. The user's operational behavior patterns are analyzed based on real-time driving status information. The intensity of steering intent is determined by the rate of change of the steering wheel angle. Acceleration, deceleration, or braking trends are identified based on the pedal operation sequence. Forward and reverse operations are distinguished based on the shift lever position and vehicle motion. User operational characteristics are matched and analyzed with standard behavioral patterns. Straight driving behavior corresponds to a stable steering wheel angle and a relatively constant accelerator pedal position. Lane changing behavior is characterized by continuous steering wheel angle changes accompanied by moderate speed adjustments. Turning behavior is characterized by a distinct steering wheel angle peak and a corresponding deceleration operation. Parking behavior is characterized by a gradual deepening of the brake pedal and a continuous decrease in speed. Reversing behavior is identified by combining gear changes and backward motion. The adaptability of each possible behavior type is evaluated based on the location distribution and attribute characteristics of the target semantic object. The probability of each behavior type occurring in the current scenario is calculated using Bayesian probabilistic reasoning, taking into account the constraints imposed by the semantic object on different behaviors and the historical statistical characteristics of user operating habits. The probability values ​​of each behavior type are compared, and the behavior type with the largest value in the probability distribution is selected as the target behavior type. When the difference between the maximum and second-largest probability values ​​is less than a preset threshold, a weighted fusion process is performed on multiple candidate behaviors. Based on the determined target behavior type, a corresponding driving intention identifier is generated. This identifier contains information such as the behavior type code, confidence value, and expected execution time window, providing clear guidance on user intention for subsequent risk assessment and intervention decisions.

[0033] In a specific embodiment, a probability distribution is calculated based on a target semantic object and possible behavior types, including: obtaining spatial geometric information of the target semantic object, performing geometric calculations and collision detection analysis based on the position distribution, shape boundary, and motion trajectory of the target semantic object to obtain a spatial feasibility area corresponding to each possible driving behavior type; collecting the user's driving state information, performing temporal feature extraction and pattern matching calculation based on the steering wheel angle change rate, pedal operation sequence, and vehicle motion state in the driving state information to obtain an operation matching weight for each possible driving behavior type; performing multi-dimensional fusion calculation based on the spatial feasibility area and the operation matching weight, performing a weighted sum operation based on the spatial constraint degree and the operation similarity to obtain a comprehensive evaluation coefficient for each possible driving behavior type; extracting the user's historical driving behavior data, and correcting the comprehensive evaluation coefficient based on the behavior selection statistics in the historical driving behavior data to obtain a behavior tendency coefficient; performing probability normalization calculation based on the behavior tendency coefficient and the interaction priority of the target semantic object to obtain a probability distribution result for each possible driving behavior type, the probability distribution result including the occurrence probability and confidence value of each behavior type.

[0034] The spatial geometric constraints of the target semantic objects are obtained. Based on the location distribution, shape boundary, and motion trajectory of each target semantic object, a corresponding action feasibility region is generated. This feasibility region defines the spatial execution range and constraints for various driving behaviors in the presence of that semantic object. By analyzing the relative relationship between the target semantic object and the vehicle's current position, the feasibility of a straight-ahead maneuver is calculated when blocked by a preceding semantic object. The safe execution space for a lane change maneuver is evaluated under the influence of a lateral semantic object. The feasible steering angle range for a turn maneuver is determined based on the distribution of semantic objects at the intersection. The boundaries of the executable region for parking and reversing maneuvers are determined under the constraints of surrounding semantic objects. The feasibility region for each possible maneuver type is determined through geometric calculations and collision detection algorithms, forming a set of spatial constraint parameters. A behavior matching weight is calculated based on the temporal characteristics of the user's current driving state information. By analyzing the time series data of steering wheel angle, pedal operation, and vehicle motion, the similarity between the user's operating pattern and the standard pattern for each possible maneuver type is quantified. The steering wheel angle change rate from driving state information is correlated with typical turning angle patterns for straight driving, lane changing, and turning. The accelerator and brake pedal operation sequences are then combined with standard operation patterns for acceleration, deceleration, and parking to assess their compatibility. Taking into account the continuity and consistency of the operation actions, short-term operation sequences are weighted, with recent operations receiving higher weights and historical operations receiving gradually lower weights. Based on the matching results, a corresponding behavior matching weight is assigned to each possible behavior type. The environmental adaptability coefficient is calculated by combining the behavior feasibility region and the behavior matching weight. By assessing the execution adaptability of each possible behavior type under the current environmental constraints, the environmental rationality of the behavior selection is quantified. The spatial extent of the behavior feasibility region is used as a spatial dimension indicator of environmental adaptability. A larger feasibility region indicates that the behavior type has better execution conditions and a higher safety margin in the current environment. The behavior matching weight is combined as a user preference dimension indicator. A higher weight indicates that the behavior type is more consistent with the user's current operational intention. The spatial dimension indicator and the user preference dimension indicator are combined through weighted multiplication to obtain an environmental adaptability coefficient that comprehensively reflects both environmental constraints and user intention. The environmental adaptability coefficient is modified by incorporating historical behavioral statistics. By analyzing the user's historical behavioral selection patterns in similar scenarios, the currently calculated environmental adaptability coefficient is personalized and optimized. Target semantic object configurations and driving state information similar to the current scenario are extracted from historical driving data. The selection frequency and execution success rate of various behavioral types under similar conditions are calculated. The historically statistically derived behavioral selection probability is used as a prior probability and combined with the current environmental adaptability coefficient as a likelihood probability through Bayesian fusion calculations to obtain a modified adaptability coefficient that takes into account the user's personalized preferences.By introducing a time decay factor to weight historical data from different periods, recent historical data has a higher reference value, while the influence of long-term data gradually decreases. The revised environmental adaptability coefficient more accurately reflects the user's actual behavioral tendencies under specific environmental conditions. Based on the revised environmental adaptability coefficient, a probability distribution for each possible behavior type is generated. Probability normalization is performed to ensure that the sum of the probabilities of all behavior types equals a standard value, forming a complete driving behavior probability distribution. The revised adaptability coefficient is used as the original probability value for each behavior type. The probability distribution is then adjusted secondary to take into account the interaction priority of the target semantic object. High-priority semantic objects have a more significant impact on the probability of the corresponding behavior type. Probability smoothing techniques are introduced to avoid extreme cases where the probabilities of certain behavior types are too low or too high, ensuring the rationality and stability of the probability distribution. The generated probability distribution is evaluated for confidence, calculating the concentration and dispersion indicators. A probability distribution with high concentration indicates a high degree of certainty in the behavior prediction, while a distribution with high dispersion indicates a wide range of possible behavior options. The resulting probability distribution contains the probability of occurrence of each possible behavior type and the corresponding confidence information, providing a quantitative probabilistic basis for subsequent target behavior type determination and driving intention identification.

[0035] In some embodiments, determining whether the vehicle is in danger based on the semantic object, the spatial calibration data and the driving intention identifier, and intervening to control the vehicle when danger occurs, includes: obtaining the corresponding safety distance threshold and collision time threshold based on the driving intention identifier; calculating the collision prediction time and predicted collision position between the vehicle and the semantic object based on the relative distance and relative speed in the spatial relationship parameters; comparing the collision prediction time with the collision time threshold to determine whether there is a collision risk; determining whether the constraint condition of the safety distance threshold is violated based on the relative distance in the spatial relationship parameters; if the collision risk exists or the constraint condition is violated, determining the danger level, the danger level including: safe, warning, dangerous and emergency; when the danger level is dangerous or emergency, generating a corresponding intervention control strategy based on the attribute information of the semantic object and the driving intention identifier and controlling the vehicle to perform risk avoidance actions.

[0036] For example, based on the target behavior type in the driving intention indicator, the corresponding safety parameters are retrieved from a preset threshold table. When the driving intention indicator indicates straight driving, a standard following safety distance threshold and a longer time-to-collision threshold are applied. Lane change or turning intentions are associated with stricter safety distance requirements and shorter time-to-collision thresholds. The predicted collision time is calculated by dividing the current relative distance between the vehicles by the absolute value of the relative closing speed. The spatial coordinates of the predicted collision location are determined based on the extrapolation of the vehicle and semantic object's motion trajectories. If the calculated predicted collision time is less than the time-to-collision threshold obtained from the threshold table, a collision risk in the temporal dimension is determined. If the relative distance in the spatial relationship parameter is less than the corresponding safety distance threshold, a constraint violation in the spatial dimension is determined. A quantitative risk assessment is performed based on the ratio of the predicted collision time to the time-to-collision threshold and the ratio of the relative distance to the safety distance threshold. Lower ratios indicate higher risk. When both ratios are greater than the safe range, the risk level is considered safe. If either ratio enters the warning range, the risk level is upgraded to a warning. Further decreases in the ratio indicate a dangerous or emergency state, respectively. When the hazard level reaches critical, a preparatory intervention control strategy is activated, including reducing the vehicle's speed and adjusting the driving trajectory to increase the safe distance from the semantic object. At the emergency level, mandatory evasive action is immediately executed, selecting either emergency braking or emergency steering based on the spatial distribution of the semantic object. For example, if a static obstacle is detected ahead and the vehicle is approaching at a high speed, emergency braking intervention control is triggered if the driving intention is indicated as going straight. However, when facing a dynamic target to the side and the driving intention is indicated as changing lanes, a steering avoidance strategy is prioritized. This differentiated intervention control strategy provides the most appropriate safety protection measures based on the specific traffic scenario and user intention.

[0037] In some embodiments, the calculation of the predicted collision time and predicted collision position between the vehicle and the semantic object based on the relative distance and relative speed in the spatial relationship parameters includes: extracting the distance component of the relative distance and the speed component of the relative speed from the spatial relationship parameters, the distance component including: longitudinal distance and lateral distance, and the speed component including: longitudinal relative speed and lateral relative speed; calculating the relative motion vector between the vehicle and the semantic object based on the longitudinal relative speed and the lateral relative speed, and determining the collision approach direction according to the relative motion vector; predicting the position coordinate sequence of the vehicle and the semantic object at a future moment based on the longitudinal distance, the lateral distance and the collision approach direction; determining the minimum distance moment and the corresponding predicted collision position between the vehicle and the semantic object based on the position coordinate sequence; and obtaining the collision prediction time based on the time difference between the minimum distance moment and the current moment.

[0038] For example, a coordinate system transformation is used to project the three-dimensional relative distance in the spatial relationship parameters onto a longitudinal distance component in the vehicle's forward direction and a lateral distance component perpendicular to the vehicle's forward direction. The three-dimensional relative velocity vector is then decomposed into a longitudinal relative velocity component along the vehicle's longitudinal axis and a lateral relative velocity component along the vehicle's lateral axis. Using vector synthesis, the longitudinal and lateral relative velocities are summed to obtain the magnitude and direction of the relative motion vector. The angle between the relative motion vector and the vehicle's longitudinal axis is calculated using the inverse tangent function to determine the collision approach direction. Based on the current longitudinal and lateral distances, as well as the collision approach direction, the future motion trajectories of the vehicle and semantic object are extrapolated using the uniform linear motion assumption, generating a sequence of position coordinates at multiple future moments. The instantaneous distance between the vehicle and the semantic object is calculated at each moment in the position coordinate sequence. The moment with the minimum distance is identified as the minimum distance moment. The vehicle and semantic object coordinates corresponding to this moment are the predicted collision location. The time difference between the minimum distance moment and the current moment is subtracted from the minimum distance moment to obtain the predicted collision time. This time value reflects the time required for the vehicle and semantic object to reach the closest distance, assuming the current motion state remains unchanged. For example, when a vehicle is chasing a semantic object ahead at a constant speed, also moving at a constant but slower speed, the longitudinal relative velocity is positive while the lateral relative velocity approaches zero. The collision direction is forward along the vehicle's longitudinal axis. Extrapolation allows for precise prediction of the time and location of the collision between the two. This prediction method provides accurate spatiotemporal reference information for timely initiation of evasive measures. The technical effect is reflected in the ability to identify potential collision hazards in advance and provide a precise time window for the formulation of intervention control strategies. This avoids the false positives or missed positives that can occur with traditional methods based solely on distance, significantly improving the accuracy and reliability of assisted driving tests.

[0039] An embodiment of the present application provides a car, which is used to execute an intervention control method for assisted driving as described in any one of the embodiments of the present application.

[0040] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An intervention control method for assisted driving, characterized in that: The method comprises: Acquire a panoramic stitched image of the test environment, and identify a set of calibration objects in the panoramic stitched image; generating a semantic object according to the set of calibration objects; Acquiring posture data of a vehicle, and determining spatial relationship parameters between the vehicle and the semantic object based on the semantic object and the posture data; performing intention classification based on the semantic object and the user's driving state information to obtain a driving intention identifier; Determine whether the vehicle is in danger based on the semantic object, the spatial calibration data and the driving intention identifier, and intervene to control the vehicle when danger occurs.

2. The intervention control method for assisted driving according to claim 1, wherein: The step of acquiring a panoramic stitched image of the test environment and identifying a set of calibration objects in the panoramic stitched image includes: capturing multi-view image data of the test environment; Registering the multi-view image data to generate a panoramic stitched image of the test environment; Identifying calibration objects in the panoramic stitched image, the calibration objects including: a grid calibration plate, circular marking points, and a linear calibration ruler; The calibration objects in the panoramic stitched image are detected to form a calibration object set.

3. The intervention control method for assisted driving according to claim 1, wherein: Generating a semantic object according to the set of calibration objects includes: Based on the calibration object set and the preset test configuration file, establishing a three-dimensional spatial coordinate system and a virtual test object of the test environment, wherein the virtual test object includes: a static obstacle, a dynamic target, and a path guide point; Performing semantic classification and annotation on the virtual test objects to determine attribute information of each virtual test object, wherein the attribute information includes: object type, motion state, and interaction priority; A semantic object is generated according to the virtual test object and the attribute information.

4. The intervention control method for assisted driving according to claim 1, wherein: The acquiring of the vehicle's posture data and determining a spatial relationship parameter between the vehicle and the semantic object according to the semantic object and the posture data includes: Extracting position state data and static attribute information of the semantic object, wherein the static attribute information includes: object size, shape outline and fixed position coordinates; Calculating a dynamic spatial relationship between the vehicle and the semantic object based on the posture data and the position state data, wherein the dynamic spatial relationship parameters include: relative distance, relative angle, and relative speed; Based on the static attribute information, calculating the spatial constraint relationship of the vehicle relative to the semantic object to obtain a static spatial relationship; A spatial relationship parameter is generated according to the dynamic spatial relationship parameter and the static spatial relationship.

5. The intervention control method for assisted driving according to claim 3, wherein: The performing intention classification according to the semantic object and the user's driving state information to obtain a driving intention identifier includes: Selecting a target semantic object having an interactive relationship with the vehicle from the semantic objects; identifying possible behavior types of the user based on the driving state information, the possible behavior types including: going straight, changing lanes, turning, parking, and reversing; Calculating a probability distribution based on the target semantic object and the possible behavior type; determining a target behavior type from the possible behavior types according to a maximum probability value in the probability distribution; A driving intention identifier is determined according to the target behavior type.

6. The intervention control method for assisted driving according to claim 5, wherein: The determining whether the vehicle is in danger based on the semantic object, the spatial calibration data, and the driving intention identifier, and intervening to control the vehicle when danger occurs, includes: According to the driving intention identifier, obtaining a corresponding safety distance threshold and collision time threshold; Calculating a predicted collision time and a predicted collision position between the vehicle and the semantic object based on a relative distance and a relative speed in the spatial relationship parameters; Comparing the collision prediction time with the collision time threshold to determine whether there is a collision risk; Determining whether the constraint condition of the safety distance threshold is violated based on the relative distance in the spatial relationship parameter; If the collision risk or constraint violation exists, determining a danger level, wherein the danger level includes: safe, warning, dangerous, and emergency; When the danger level is dangerous or urgent, a corresponding intervention control strategy is generated according to the attribute information of the semantic object and the driving intention identifier, and the vehicle is controlled to perform a risk avoidance action.

7. The intervention control method for assisted driving according to claim 6, wherein: The calculating, based on the relative distance and relative speed in the spatial relationship parameters, the predicted collision time and predicted collision position between the vehicle and the semantic object comprises: Extracting the distance component of the relative distance and the speed component of the relative speed from the spatial relationship parameters, wherein the distance component includes: longitudinal distance and lateral distance, and the speed component includes: longitudinal relative speed and lateral relative speed; calculating a relative motion vector between the vehicle and the semantic object based on the longitudinal relative speed and the lateral relative speed, and determining a collision approach direction according to the relative motion vector; predicting a position coordinate sequence of the vehicle and the semantic object at a future time based on the longitudinal distance, the lateral distance, and the collision approach direction; Determining a minimum distance moment between the vehicle and the semantic object and a corresponding predicted collision position based on the position coordinate sequence; The collision prediction time is obtained according to the time difference between the minimum distance moment and the current moment.

8. An automobile, characterized in that: The automobile is used to execute the intervention control method for assisted driving as described in any one of claims 1 to 7.

Images