Vehicle control method, electronic device, and computer-readable storage medium

Abstract

Embodiments of the present application provide a vehicle control method, an electronic device and a computer readable storage medium, wherein the method comprises: in response to detecting that a vehicle is in a preset curve scene, obtaining vehicle trajectory data and visual perception images of the vehicle; based on a plurality of dynamic indicators, respectively extracting features of the vehicle trajectory data and the visual perception images to obtain vehicle state features and visual perception features corresponding to the plurality of dynamic indicators respectively, wherein different dynamic indicators are used to represent state quantities of different path features of the vehicle driving on a road; performing consistency evaluation on the vehicle state features and the visual perception features to obtain a target evaluation value; based on the target evaluation value, performing weighted fusion on the vehicle trajectory data and the visual perception images to obtain a target fusion result, and controlling the vehicle based on the target fusion result. The present application solves the technical problem of poor vehicle control accuracy in specific scenes such as strong light and high dynamic curve in the related art.

Description

Technical Field

[0001] This application relates to the field of vehicle autonomous driving control technology, and in particular to a vehicle control method, electronic device, and computer-readable storage medium. Background Technology

[0002] In the field of intelligent driving vehicle technology, multi-sensor fusion systems have become a core means to achieve high-precision environmental perception and are widely used in Level 2 Plus (L2+) autonomous driving scenarios. Current mainstream solutions in the industry mostly employ fusion strategies, confidence assessment methods based on sensor internal quality indicators, or rely on Dempster-Shafer Theory (DS) evidence theory for general scenarios for decision fusion. These methods cannot distinguish between spurious sensor outputs caused by environmental interference and deviations from actual motion constraints. This leads to abrupt changes in fusion results and control instability in specific scenarios such as strong light and highly dynamic curves, resulting in poor vehicle control accuracy in these conditions. Summary of the Invention

[0003] This application provides a vehicle control method, electronic device, and computer-readable storage medium, aiming to improve the problem of poor vehicle control accuracy in specific scenarios such as strong light and highly dynamic curves in related technologies.

[0004] According to one aspect of the embodiments of this application, a vehicle control method is provided, the method comprising: in response to detecting that a vehicle is in a preset curved scenario, acquiring vehicle trajectory data and visual perception images of the vehicle; extracting features from the vehicle trajectory data and visual perception images based on multiple dynamic indicators to obtain vehicle state features and visual perception features corresponding to the multiple dynamic indicators, wherein different dynamic indicators are used to represent state quantities of different path features of the vehicle traveling on the road; performing a consistency evaluation on the vehicle state features and visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and visual perception features on multiple dynamic indicators; performing a weighted fusion of the vehicle trajectory data and visual perception images based on the target evaluation value to obtain a target fusion result, and controlling the vehicle based on the target fusion result.

[0005] In this embodiment of the invention, in response to detecting that a vehicle is in a preset curve scenario, vehicle trajectory data and visual perception images are acquired; features are extracted from the vehicle trajectory data and visual perception images based on multiple dynamic indicators to obtain vehicle state features and visual perception features corresponding to the multiple dynamic indicators, wherein different dynamic indicators are used to represent state quantities of different path features of the vehicle traveling on the road; consistency evaluation is performed on the vehicle state features and visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and visual perception features on multiple dynamic indicators; the vehicle trajectory data and visual perception images are weighted and fused based on the target evaluation value to obtain a target fusion result, and the vehicle is controlled based on the target fusion result. This application embodiment is based on a trajectory and visual consistency evaluation mechanism with multiple dynamic indicators. By extracting features from vehicle trajectory data and visual perception images, it obtains vehicle state features and visual perception features corresponding to multiple dynamic indicators. Furthermore, it performs consistency evaluation on the vehicle state features and visual perception features to obtain a target evaluation value, that is, to obtain the difference between the visual perception image and the vehicle trajectory data. Based on the target evaluation value, it achieves the purpose of quantifying the reliability of the visual perception image, thereby realizing the technical effect of controlling the vehicle based on the reliability of the visual perception image. This solves the technical problem of poor vehicle control accuracy in specific scenarios such as strong light and high dynamic curves in related technologies.

[0006] Furthermore, the vehicle trajectory data and visual perception images are weighted and fused based on the target evaluation value to obtain the target fusion result, and the vehicle is controlled based on the target fusion result, including: determining the visual weight value corresponding to the visual perception image based on the preset correspondence and the target evaluation value; determining the state weight value corresponding to the vehicle trajectory data based on the difference between the preset value and the visual weight value; and weighting and fusing the visual perception image and the vehicle trajectory data based on the visual weight value and the state weight value to obtain the target fusion result.

[0007] By dynamically adjusting the weights of visual and trajectory data, a smooth switching of perception sources can be achieved, avoiding control abrupt changes caused by visual failure, improving the stability of the fusion results and the robustness of the system, and ensuring safe and continuous control of the vehicle in high-risk scenarios.

[0008] Furthermore, based on the weight values ​​of multiple dynamic indicators, a consistency assessment is performed on vehicle state features and visual perception features to obtain a target evaluation value, including: determining the feature difference values ​​corresponding to multiple dynamic indicators based on vehicle state features and visual perception features; and weighted fusion of the feature difference values ​​corresponding to multiple dynamic indicators based on the weight values ​​of multiple dynamic indicators to obtain the target evaluation value.

[0009] By weighted fusion of the feature differences of multiple dynamic indicators, a quantitative assessment of the consistency between vision and trajectory is achieved, improving the physical credibility and scene adaptability of the assessment results and providing an accurate basis for subsequent dynamic adjustment of weights.

[0010] Furthermore, the method also includes: acquiring the angular velocity and acceleration collected by the inertial measurement unit in the vehicle; integrating the angular velocity and acceleration to obtain the vehicle's position information, vehicle velocity information, and vehicle attitude information; and predicting the vehicle's trajectory based on the vehicle position information, vehicle velocity information, and vehicle attitude information to obtain vehicle trajectory data.

[0011] Based on inertial measurement units and high-precision map information, continuous trajectory estimation without visual dependence is achieved, improving the stability and reliability of vehicle pose estimation in visual failure scenarios.

[0012] Furthermore, based on vehicle location information, vehicle speed information, and vehicle attitude information, vehicle trajectory data is determined, including: obtaining map information of the road where the vehicle is currently located; predicting the vehicle trajectory based on the vehicle location information, vehicle speed information, and vehicle attitude information to obtain initial vehicle trajectory data; constraining the location information in the initial vehicle trajectory data to the road centerline corresponding to the map information, and constraining the attitude information in the initial vehicle trajectory data to the tangent direction of the road centerline to obtain vehicle trajectory data.

[0013] By combining inertial navigation calculations with high-precision map constraints, the cumulative error of the inertial measurement unit is effectively suppressed, the long-term accuracy and geometric rationality of trajectory data are improved, and a highly reliable reference benchmark is provided for visual consistency assessment.

[0014] Furthermore, the method also includes: acquiring the vehicle's location information and the road geometry information of the road where the vehicle is currently located; and determining whether the vehicle is in a preset curve scenario based on the vehicle's location information and the road geometry information.

[0015] Based on real-time location and high-precision map road geometry information, high-risk scenarios such as tunnel exit curves can be accurately identified, enabling early prediction of perceived risks and providing a time-series triggering basis for subsequent dynamic fusion strategies.

[0016] Furthermore, based on vehicle location information and road geometry information, it is determined whether the vehicle is in a preset curve scenario, including: determining the target distance between the vehicle's current position and the preset position based on map information; determining the road curvature at the pre-aiming point based on road geometry information; determining the target heading angle difference based on the first heading angle of the current position and the second heading angle of the target position on the road, wherein the target position is a position on the road at a preset distance from the current position; and determining whether the vehicle is in a preset curve scenario based on the target distance, road curvature, and target heading angle difference.

[0017] By jointly judging multi-dimensional geometric parameters, the system can accurately identify sharp bends at tunnel exits, avoid false triggering by a single condition, and improve the robustness and real-time performance of scene recognition.

[0018] Furthermore, based on the target distance, road curvature, and target heading angle difference, it is determined whether the vehicle is in a preset curve scenario, including: in response to the target distance being less than a preset distance threshold, the road curvature being less than a preset curvature threshold, and the target heading angle difference being greater than a preset heading angle threshold, it is determined that the vehicle is in a preset curve scenario; in response to the target distance being greater than or equal to a preset distance threshold, or the road curvature being greater than or equal to a preset curvature threshold, or the target heading angle difference being less than or equal to a preset heading angle threshold, it is determined that the vehicle is not in a preset curve scenario.

[0019] By employing a triple threshold logic linkage judgment, the system accurately identifies complex scenarios involving sharp bends at tunnel exits, effectively avoiding false triggers and missed triggers, and improving the accuracy and reliability of scene recognition.

[0020] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory for storing a computer program; and a processor for executing the program stored in the memory, wherein the program executes the above-described method when it runs.

[0021] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein the executable program controls the device where the computer-readable storage medium is located to perform the above-described method when it runs. Attached Figure Description

[0022] Figure 1 This is a flowchart of a vehicle control method provided in an embodiment of this application;

[0023] Figure 2 This is a schematic diagram of tunnel exit curve scene recognition provided in an embodiment of this application;

[0024] Figure 3 This is a schematic diagram of theoretical trajectory and feature generation based on dynamic consistency provided in an embodiment of this application;

[0025] Figure 4 This is a schematic diagram of visual perception feature extraction and dynamic consistency index calculation provided in an embodiment of this application;

[0026] Figure 5 This is a schematic diagram of dynamic arbitration and fusion of multi-sensor confidence based on consistency index provided in an embodiment of this application;

[0027] Figure 6 This is a schematic diagram of a vehicle control device provided in an embodiment of this application;

[0028] Figure 7 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0029] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0030] This application provides a vehicle control method. Figure 1 This is a flowchart of a vehicle control method provided in an embodiment of this application, such as... Figure 1 As shown, it includes the following steps:

[0031] Step S102: In response to detecting that the vehicle is in a preset curve scenario, acquire the vehicle trajectory data and visual perception image.

[0032] The aforementioned preset curve scenarios refer to a type of working condition with specific perceptual challenges, predefined based on prior geographical information and road geometry features of the vehicle's environment. The determination of preset curve scenarios includes a combination of multi-dimensional conditions such as vehicle spatial position, road curvature ahead, heading abrupt change rate, or ambient lighting abrupt change, used to trigger the system to enter a high-risk perception enhancement mode and activate specific sensor fusion strategies. These preset curve scenarios can include sharp bends at tunnel exits, or scenarios where visual sensors fail due to blindness caused by strong light or sudden changes in field of view.

[0033] The aforementioned preset curve scenario can be a sharp bend at a tunnel exit, where the road curvature at the pre-aiming point ahead is less than a preset curvature threshold. The aforementioned preset curve scenario can also be a scenario where the visual sensor fails due to blindness caused by strong light or a sudden change in field of view.

[0034] The pre-defined curve scenario recognition mechanism integrates prior geographical location knowledge with real-time road geometry features to proactively predict and trigger high-risk driving scenarios. Specifically, it first obtains the vehicle's real-time latitude and longitude coordinates using a high-precision positioning module and combines this with pre-stored tunnel structure topology data in a high-precision map to calculate the relative spatial distance between the vehicle and the tunnel exit. If the vehicle enters a pre-defined threshold range near the tunnel exit, it is determined that the vehicle is in this specific geographical area. Simultaneously, it extracts the set of road centerline points within a pre-defined look-ahead distance behind the tunnel exit from the high-precision map, calculates the road curvature using a curve fitting algorithm, and then derives the radius of curvature for this road segment. If the calculated radius of curvature is less than a pre-defined curvature threshold, the road segment is determined to be a sharp curve scenario. If both conditions are met—the vehicle being located near the tunnel exit and the road exhibiting high curvature characteristics after the exit—the vehicle is confirmed to have entered a pre-defined curve scenario.

[0035] The aforementioned vehicle trajectory data refers to the spatiotemporal state sequence of a vehicle during its movement in space, calculated jointly or directly acquired by sensors or data sources such as Inertial Measurement Units (IMUs), Global Navigation Satellite Systems (GNSS), and high-precision maps. Vehicle trajectory data can include dynamic parameters such as the vehicle's position, speed, heading angle, and rate of change. Vehicle trajectory data serves as a reference benchmark for evaluating the reliability of visual perception.

[0036] The aforementioned visual perception images refer to the environmental optical image data collected by vehicle-mounted cameras. After image preprocessing and feature extraction, these images are used to identify road structures, obstacles, or environmental semantic information—either raw or processed image streams. The content of visual perception images is affected by external environmental factors such as lighting, occlusion, and lens glare, making them a primary but easily malfunctioning source of perception input in autonomous driving systems. The aforementioned road structures can include lane lines, curbs, etc.

[0037] In one alternative embodiment, in response to detecting that the vehicle is in a preset curve scenario, the vehicle control system generates vehicle trajectory data by fusing high-precision map and inertial navigation data, and simultaneously acquires visual perception images from the on-board vision sensor for subsequent weighted fusion.

[0038] In another alternative embodiment, in response to detecting that the vehicle is in a preset curve scenario, the vehicle control system calculates the vehicle's trajectory using a tightly coupled filtering algorithm by fusing high-frequency attitude and velocity data from the inertial measurement unit and combining it with road geometric constraints provided by a high-precision map. This generates highly reliable vehicle trajectory data. Simultaneously, visual perception images are acquired by the onboard vision sensor as visual perception input.

[0039] The aforementioned high-frequency attitude and velocity data refers to the instantaneous angular velocity and linear acceleration of the vehicle output by the inertial measurement unit at a high sampling frequency. After integration processing, the real-time attitude and velocity changes of the vehicle can be calculated. The aforementioned road geometric constraints refer to the structured geometric parameters pre-stored in the high-precision map, such as road centerline curvature, slope, heading angle, lane width, and lateral offset, used to constrain vehicle motion to conform to the actual road topology and suppress navigation error drift. The aforementioned tightly coupled filtering algorithm is a multi-sensor fusion method that directly inputs the raw observations from the inertial sensors, such as acceleration and angular velocity, along with the prior position and orientation information provided by the map, into a unified state estimation algorithm.

[0040] The embodiments of this application realize synchronous acquisition of sensor input and scene adaptive triggering, providing a reliable and synchronous multi-source data foundation for subsequent dynamic adjustment of confidence, and improving the response timeliness and decision accuracy of the vehicle control system in high-risk environments.

[0041] Step S104: Based on multiple dynamic indicators, feature extraction is performed on vehicle trajectory data and visual perception images to obtain vehicle state features and visual perception features corresponding to multiple dynamic indicators. Among them, different dynamic indicators are used to represent the state quantities of different path features of the vehicle traveling on the road.

[0042] The aforementioned dynamic indicators refer to physical quantities defined based on vehicle kinematics and road geometric constraints, used to quantify key state changes of a vehicle during operation, such as lateral offset, heading deviation angle, and road curvature matching degree. Essentially, dynamic indicators are calculable parameters reflecting the dynamic consistency between the vehicle's actual motion state and the road environment, used to construct objective and quantifiable assessment criteria for perceived reliability.

[0043] The aforementioned vehicle state characteristics refer to the physical quantities extracted from vehicle trajectory data that directly correspond to various dynamic indicators, such as the vehicle's lateral deviation from the road center, the angle between the actual heading and the road heading, and the road curvature response ahead of the vehicle's current position. Vehicle state characteristics are used to characterize the vehicle's true motion state in physical space.

[0044] The aforementioned visual perception features refer to the observation values ​​extracted from visually perceived images that correspond to the same dynamic indicators, such as the lateral position of lane lines obtained from visual recognition, the angle between the fitted lane direction and the vehicle's heading, and the estimated road curvature extracted from the image. Visual perception features reflect the visual system's perception of the road environment and are easily affected by environmental interference, leading to biases.

[0045] The aforementioned state quantities representing different path characteristics refer to physical quantities used to quantify several key kinematic and geometric properties exhibited by a vehicle along its driving path in road space, such as lateral offset, heading deviation angle, and curvature response. Each type of state quantity corresponds to a specific structural feature of a road path, such as straight sections, curves, and curvature variation zones. Essentially, the state quantities representing different path characteristics are measurable dimensions describing the relationship between the vehicle's trajectory and road geometric constraints, forming an important physical basis for evaluating perceptual consistency.

[0046] In one optional embodiment, the vehicle control system extracts dynamic indices from the vehicle trajectory data by calculating the vertical distance between the vehicle's centerline and the road reference path, the angle between the vehicle's longitudinal axis and the road tangent, and the second derivative response of the trajectory curve. For the visually perceived image, the vehicle control system employs edge detection and lane line fitting algorithms to extract the corresponding lateral position in the world coordinate system, the visual heading angle, and the curvature of the fitted curve, forming a visual feature vector of the same dimension as the trajectory data.

[0047] In one optional embodiment, the vehicle control system first extracts vehicle state features corresponding to dynamic indices from the trajectory generated by the fusion of inertial navigation and high-precision maps. Vehicle state features include, but are not limited to, lateral offset, heading angle deviation, yaw rate, trajectory curvature, and velocity-curvature product. Simultaneously, the vehicle control system extracts corresponding visual perception features from the visually perceived images using image processing and deep learning algorithms. Visual perception features include, but are not limited to, lane line lateral position estimation in the vehicle coordinate system, fitted lane heading trend, local road curvature estimation, lateral acceleration estimation, and perceived lane width. These features are parsed by the visual sensor and mapped to the vehicle motion coordinate system through coordinate transformation. Subsequently, the vehicle control system aligns each pair of vehicle state features with the visual perception features item by item, forming feature pairs across multiple dynamic dimensions, i.e., multiple dynamic indices.

[0048] The aforementioned dynamic parameters can include lateral offset, heading angle deviation, trajectory curvature, lateral acceleration deviation, longitudinal acceleration deviation, and vehicle yaw rate deviation, etc.

[0049] The aforementioned lateral offset refers to the vertical distance between the vehicle's center and the road centerline, reflecting whether the vehicle is centered. The aforementioned heading angle deviation refers to the angle between the vehicle's forward direction and the actual road direction, reflecting whether the vehicle is turning in the expected direction. The aforementioned trajectory curvature refers to the degree of curvature of the vehicle's path; the greater the curvature, the sharper the turn. The aforementioned edge detection is a technique that identifies object contour boundaries through image grayscale changes, used to find the light and dark boundary areas of lane lines. The aforementioned lane line fitting uses a mathematical curve to fit the detected discrete lane line pixels, obtaining a continuous and smooth lane line model used to estimate the path shape and the relative position of vehicles.

[0050] In another optional embodiment, the vehicle control system calculates three types of dynamic path features—longitudinal acceleration rate of change, lateral acceleration, and steering angle response—based on the kinematic model of the vehicle trajectory data. For visually perceived images, the vehicle control system uses a deep learning model to output the three-dimensional spatial parameters of the lane lines and converts them into relative offset, viewpoint change angle, and local curvature gradient in the vehicle coordinate system. Spatiotemporal alignment ensures a one-to-one correspondence between visual features and trajectory features in physical semantics, achieving multi-dimensional dynamic consistency modeling.

[0051] The kinematic model of the aforementioned vehicle trajectory data refers to a mathematical model describing the relationship between vehicle position, speed, steering, and time. Based on vehicle geometry and motion laws, it is used to predict the vehicle's future path. The longitudinal acceleration rate of change refers to how quickly the vehicle's acceleration in the longitudinal direction changes over time, reflecting the smoothness of acceleration or braking. The lateral acceleration refers to the lateral acceleration caused by the curvature of the path when the vehicle turns, reflecting the degree of steering aggression. The steering angle response refers to the delay and magnitude of the change in the vehicle's actual heading after a change in steering wheel angle, reflecting the vehicle's dynamic steering characteristics.

[0052] The aforementioned deep learning model refers to an intelligent algorithm that automatically learns features from image data through multi-layer neural networks. It can recognize complex patterns such as lane lines, vehicles, and signs. The deep learning model takes images captured by an onboard camera as input, extracts image features through a convolutional network, and combines these features with known camera intrinsic and extrinsic parameters, such as mounting height, tilt angle, and focal length, to directly predict the three-dimensional spatial point sequence of lane lines in the vehicle coordinate system, or output the polynomial coefficients of the lane lines, such as cubic polynomial parameters, thereby reconstructing the spatial shape and position of the vehicle in the real world.

[0053] This application embodiment achieves feature extraction of vehicle state characteristics and visual perception characteristics by acquiring multi-dimensional dynamic indicators, constructs a quantifiable evaluation model based on physical laws, and improves the objectivity and accuracy of perception reliability judgment.

[0054] Step S106: Perform a consistency assessment on the vehicle state features and visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and visual perception features on multiple dynamic indicators.

[0055] The aforementioned target evaluation value refers to a comprehensive difference metric generated by comparing and weighting vehicle state characteristics and visual perception characteristics under multiple dynamic indicators. It objectively characterizes the consistency level between the visual perception result and the actual physical trajectory in multi-dimensional path features. The magnitude of the target evaluation value directly reflects the current reliability status of the vision system and serves as input for the dynamic adjustment of sensor confidence. A smaller target evaluation value indicates a more reliable visual perception image, while a larger target evaluation value indicates more reliable vehicle trajectory data.

[0056] The aforementioned consistency assessment refers to the quantitative comparison between the perception output of a sensor (such as vision) and an independent, highly reliable reference benchmark constructed based on physical laws or prior knowledge, such as a theoretical trajectory generated by a combination of an inertial navigation system and a high-precision map. The degree of deviation between the two in key state quantities, such as position, heading, and curvature, is calculated to objectively assess the credibility and physical rationality of the sensor's perception results in the current environment and scenario.

[0057] In one optional embodiment, the vehicle control system calculates the absolute difference between the vehicle state characteristics and the visual perception characteristics for each dynamic index, and then weights and sums the differences of each index according to preset weights to obtain a target evaluation value. The smaller the target evaluation value, the more consistent the visual perception result is with the actual motion state of the vehicle. The larger the target evaluation value, the more interference or anomaly exists in the vision system. The evaluation value serves as the quantitative basis for subsequent confidence adjustment.

[0058] In another alternative embodiment, the vehicle control system inputs vehicle state features and visual perception features into a physical constraint-based training contrast network. This network learns the distribution patterns of these two features in a multidimensional dynamic space and outputs a normalized similarity score. This similarity score reflects the consistency level of the vehicle's overall motion behavior and serves as the target evaluation value.

[0059] The contrast network trained under the aforementioned physical constraints is a type of neural network. During training, this network not only uses image and sensor data but also forcibly incorporates the physical laws of vehicle kinematics and road geometry, such as vehicles not being allowed to leave their lanes and the need for corresponding lateral acceleration when turning. By comparing the actual trajectory with the visual perception result, the contrast network learns that the two should be similar when conforming to physical laws, and judges them as inconsistent when they deviate from physical laws, thus outputting a consistency score reflecting whether the consistency is reasonable.

[0060] This application embodiment constructs a visual reliability quantification mechanism based on physical constraints through multi-dimensional feature consistency evaluation, enabling the vehicle control system to accurately perceive the vehicle's motion trend and providing a reliable and quantifiable decision basis for subsequent fusion.

[0061] Step S108: The vehicle trajectory data and visual perception image are weighted and fused based on the target evaluation value to obtain the target fusion result, and the vehicle is controlled based on the target fusion result.

[0062] The aforementioned target fusion result refers to the final environmental state estimate obtained by linear or nonlinear weighting calculation based on the target evaluation value, i.e., their respective dynamic confidence weights, after the environmental information represented by the visual perception image and the kinematic true value represented by the vehicle trajectory data.

[0063] In one optional embodiment, the confidence level of the visually perceived image is dynamically weighted based on the target evaluation value. If the target evaluation value indicates that the visual perception is unreliable, the vehicle control system generates control commands primarily based on vehicle trajectory data, with the visually perceived image serving only as an auxiliary correction input to ensure stable control output, generating a target fusion result, and further controlling the vehicle based on the target fusion result. If the target evaluation value indicates that the visual perception is reliable, the vehicle control system continues to generate control commands primarily based on the visually perceived image, generating a target fusion result, and further controlling the vehicle based on the target fusion result.

[0064] In another optional embodiment, the vehicle control system smoothly adjusts the fusion ratio of vehicle trajectory data and visually perceived images in the control algorithm based on the continuous changing trend of the target evaluation value to generate the target fusion result. In this embodiment, steering, acceleration, and deceleration commands can be gradually adjusted according to the perceived reliability, avoiding abrupt changes in control commands.

[0065] The embodiments of this application realize a closed loop of perception reliability and control decision-making, and adaptively fuse multi-source information based on target evaluation values ​​to improve the stability and safety of the control system in the event of abnormal perception.

[0066] In this embodiment of the invention, in response to detecting that a vehicle is in a preset curve scenario, vehicle trajectory data and visual perception images are acquired; features are extracted from the vehicle trajectory data and visual perception images based on multiple dynamic indicators to obtain vehicle state features and visual perception features corresponding to the multiple dynamic indicators, wherein different dynamic indicators are used to represent state quantities of different path features of the vehicle traveling on the road; consistency evaluation is performed on the vehicle state features and visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and visual perception features on multiple dynamic indicators; the vehicle trajectory data and visual perception images are weighted and fused based on the target evaluation value to obtain a target fusion result, and the vehicle is controlled based on the target fusion result. This application embodiment is based on a trajectory and visual consistency evaluation mechanism with multiple dynamic indicators. By extracting features from vehicle trajectory data and visual perception images, it obtains vehicle state features and visual perception features corresponding to multiple dynamic indicators. Furthermore, it performs consistency evaluation on the vehicle state features and visual perception features to obtain a target evaluation value, that is, to obtain the difference between the visual perception image and the vehicle trajectory data. Based on the target evaluation value, it achieves the purpose of quantifying the reliability of the visual perception image, thereby realizing the technical effect of controlling the vehicle based on the reliability of the visual perception image. This solves the technical problem of poor vehicle control accuracy in specific scenarios such as strong light and high dynamic curves in related technologies.

[0067] Optionally, the vehicle trajectory data and visual perception images are weighted and fused based on the target evaluation value to obtain a target fusion result, and the vehicle is controlled based on the target fusion result, including: determining the visual weight value corresponding to the visual perception image based on a preset correspondence and the target evaluation value; determining the state weight value corresponding to the vehicle trajectory data based on the difference between the preset value and the visual weight value; and weighting and fusing the visual perception image and the vehicle trajectory data based on the visual weight value and the state weight value to obtain the target fusion result.

[0068] The aforementioned predefined correspondence refers to the functional mapping relationship between the predefined target evaluation value and the visual weight value. It is usually a monotonically decreasing continuous function, such as the logistic function (S-shaped function) or linear decay, used to transform the consistency evaluation result into a quantitative value of visual perception confidence. This ensures that the higher the evaluation value, the greater the difference and the lower the visual weight, thereby realizing the dynamic and continuous allocation of the credibility of the perception source.

[0069] The aforementioned visual weight values ​​refer to normalized coefficients calculated based on the target evaluation values ​​using a preset correspondence, used to characterize the relative importance of visually perceived images in fusion decisions. The visual weight values ​​range from 0 to 1, reflecting the current availability and reliability level of the vision system, and are key parameters for achieving adaptive fusion of multi-source perception.

[0070] The aforementioned preset values ​​refer to constants or reference values ​​that are pre-set based on engineering experience, physical laws, or system requirements before system design or algorithm implementation. These values ​​serve as the basis for calculation, comparison, or decision-making. During operation, these preset values ​​remain fixed or are adjusted only within a limited range, without dynamically changing with real-time input. They are used to construct weight relationships, threshold judgments, or normalization references, ensuring the stability and predictability of system behavior.

[0071] The aforementioned state weight values ​​characterize the degree to which vehicle trajectory data dominates the fusion process. The state weight values ​​are adjusted inversely to the visual weight values. For example, the state weight value is the complementarity coefficient obtained by subtracting the visual weight value from 1. This operation ensures that the sum of the weights of the two perception sources, visual and trajectory, is always 1.

[0072] The aforementioned target fusion result refers to the comprehensive perception output generated by weighting and superimposing visual perception images and vehicle trajectory data based on visual weight values ​​and state weight values. The fusion result combines the semantic richness of vision with the stability of trajectory data, and as the input for the final control decision, it has high robustness and continuity.

[0073] In one optional embodiment, the vehicle control system calculates the visual weight value corresponding to the visually perceived image based on the target evaluation value and a preset nonlinear mapping relationship. The preset nonlinear mapping relationship ensures that the lower the target evaluation value, the higher the visual weight, and vice versa. Subsequently, the state weight value corresponding to the vehicle trajectory data is derived from the difference between the visual weight value and the preset value, ensuring that the sum of the two weights is always one. Finally, the vehicle control system multiplies the visually perceived image and the vehicle trajectory data by their respective weights and then linearly superimposes them to obtain a smooth and adaptive target fusion result.

[0074] This application embodiment achieves the fusion of vehicle trajectory data and visual perception images through a weighted complementary mechanism, thereby improving the stability of the control system.

[0075] Optionally, based on the weight values ​​of multiple dynamic indicators, a consistency assessment is performed on vehicle state features and visual perception features to obtain a target evaluation value, including: determining the feature difference values ​​corresponding to multiple dynamic indicators based on vehicle state features and visual perception features; and weighting and fusing the feature difference values ​​corresponding to multiple dynamic indicators based on the weight values ​​of multiple dynamic indicators to obtain the target evaluation value.

[0076] The weight values ​​of the aforementioned multiple dynamic indicators refer to pre-set relative importance coefficients for different dynamic indicators, such as lateral offset, heading deviation, and road curvature matching, used to reflect the contribution weight of each indicator in assessing the reliability of visual perception. The weight values ​​of multiple dynamic indicators are set based on the stability, sensitivity, and physical correlation of each indicator under typical operating conditions, ensuring that the assessment process focuses on path feature dimensions that have a greater impact on driving safety.

[0077] The aforementioned feature difference value refers to the numerical deviation between the vehicle state characteristics and the visual perception characteristics under each dynamic index. It quantifies the degree of deviation between visual perception and the actual motion state and is the basic unit for evaluating the reliability of visual perception. A larger feature difference value indicates lower reliability of vision in that feature dimension. A smaller feature difference value indicates higher reliability of vision in that feature dimension.

[0078] In one optional embodiment, multiple dynamic indicators are extracted based on vehicle state characteristics and visual perception characteristics, including but not limited to lateral offset, heading deviation, road curvature matching degree, and acceleration response consistency. For each of these indicators, the corresponding feature difference value is obtained by calculating the difference between the vehicle state characteristics and the visual perception characteristics. The feature difference value reflects the estimation error of the visual perception system in the dynamic dimension. Subsequently, the vehicle control system assigns corresponding weight coefficients to each dynamic indicator according to its relative importance to driving safety and stability, multiplies all feature difference values ​​by their corresponding weights, and then performs a weighted sum to finally generate a target evaluation value representing overall consistency.

[0079] This application embodiment achieves accurate and physically interpretable assessment of visual perception reliability through weighted multi-dimensional dynamic difference fusion, thereby improving the accuracy and scene adaptability of the assessment results.

[0080] Optionally, the method further includes: acquiring the angular velocity and acceleration collected by the inertial measurement unit in the vehicle; integrating the angular velocity and acceleration to obtain the vehicle position information, vehicle velocity information and vehicle attitude information; and predicting the vehicle trajectory based on the vehicle position information, vehicle velocity information and vehicle attitude information to obtain vehicle trajectory data.

[0081] The aforementioned inertial measurement unit (IMU) refers to a miniature sensing device that integrates an accelerometer and a gyroscope, used to continuously collect the specific force of a vehicle in three axes, namely linear acceleration and angular velocity information. The output of the IMU is a high-frequency raw motion signal, which is independent of the external environment, has a high response speed, and is the original data source for calculating the dynamic state of the vehicle.

[0082] The angular velocity mentioned above refers to the physical quantity of how fast and in which a vehicle rotates around its own coordinate axis, and is usually measured by a gyroscope in an inertial measurement unit. Angular velocity characterizes the vehicle's turning rate in space, such as the rate of change of yaw angle during turning, and the rotational speed of pitch or roll, and is a core input for calculating changes in vehicle attitude. The aforementioned coordinate axis can be the horizontal axis, vertical axis, or other similar axes.

[0083] The acceleration mentioned above refers to the physical quantity that indicates the rate and direction of change of a vehicle's velocity in space. It is typically measured by accelerometers in an inertial measurement unit and includes linear acceleration components along the vehicle's three axes. Acceleration reflects the inertial forces acting on a vehicle during acceleration, deceleration, or cornering, and is the fundamental basis for calculating changes in vehicle speed and position. The three axes mentioned above can be longitudinal, lateral, and vertical.

[0084] The vehicle position information mentioned above refers to the absolute or relative coordinates of the vehicle in three-dimensional space obtained by integrating the acceleration collected by the inertial measurement unit over time. It reflects the spatial displacement of the vehicle over time and is a state parameter for constructing the vehicle's motion trajectory.

[0085] The vehicle speed information mentioned above refers to the instantaneous velocity and direction of the vehicle obtained by integrating the acceleration collected by the inertial measurement unit, representing the vehicle's dynamic behavior in space. The accuracy of the vehicle speed information directly affects the cumulative error of position estimation and requires map constraints to achieve long-term stability.

[0086] The aforementioned vehicle attitude information refers to the pitch, roll, and heading angles of the vehicle in three-dimensional space, obtained through the angular velocity of the integral inertial measurement unit. These angles describe the vehicle's orientation relative to the road coordinate system and are key inputs for calculating the geometric matching relationship between the vehicle and the road.

[0087] In one optional embodiment, the vehicle control system acquires the vehicle's three-axis angular velocity and linear acceleration data through the vehicle's onboard inertial measurement unit, and simultaneously obtains high-precision geometric information of the current road. Subsequently, the vehicle control system integrates the angular velocity over time to calculate the vehicle's attitude change sequence in space, including but not limited to heading, pitch, and roll angles. The vehicle control system integrates the acceleration over time, and, combined with the initial velocity conditions, gradually reconstructs the vehicle's instantaneous velocity and position changes. Finally, the vehicle control system jointly constructs continuous vehicle motion trajectory data from the calculated position, velocity, and attitude information, serving as a highly reliable motion reference independent of visual perception for subsequent perception consistency assessment and multi-source fusion decision-making.

[0088] The embodiments of this application construct a highly reliable motion benchmark that does not rely on vision. By fusing inertial calculations with map constraints, highly reliable vehicle trajectory data is generated, providing a physical basis for visual reliability assessment.

[0089] Optionally, vehicle trajectory data is determined based on vehicle position information, vehicle speed information, and vehicle attitude information, including: obtaining map information of the road where the vehicle is currently located; predicting the vehicle trajectory based on the vehicle position information, vehicle speed information, and vehicle attitude information to obtain initial vehicle trajectory data; constraining the position information in the initial vehicle trajectory data to the road centerline corresponding to the map information, and constraining the attitude information in the initial vehicle trajectory data to the tangent direction of the road centerline to obtain vehicle trajectory data.

[0090] The aforementioned map information refers to road geometry data provided by high-precision maps. Map information includes, but is not limited to, lane centerline curvature, heading change trends, lateral boundaries, and geographic coordinate references, which are used to provide prior road constraints for vehicle motion and serve as external benchmarks for inertial navigation error correction and accurate trajectory constraints.

[0091] The aforementioned initial vehicle trajectory data refers to the original trajectory sequence calculated by kinematic integration and prediction model based on the vehicle position, velocity, and attitude information output by the inertial measurement unit. It reflects the vehicle's motion trend without external constraints and has high-frequency response characteristics. However, due to the inherent drift of the inertial sensor, there is accumulated error, which requires external reference correction to improve long-term accuracy.

[0092] In one optional embodiment, the vehicle control system first acquires map information of the road where the vehicle is currently located from a high-precision map. This map information includes centerline data and tangent direction information. Then, using the vehicle's current position, speed, and attitude information, and combining this with the vehicle's kinematic model, it extrapolates to obtain initial vehicle trajectory data containing a sequence of position and attitude. Next, the positions of each point in the initial vehicle trajectory data are projected or modified onto the corresponding road centerline on the map to eliminate positioning drift errors. Simultaneously, the heading angle in the initial vehicle trajectory data is constrained to the tangent direction of the road centerline at the corresponding position to eliminate heading divergence errors. Finally, vehicle trajectory data with high consistency and high reliability, corrected by map geometric constraints, is output.

[0093] The embodiments of this application improve the accuracy and stability of trajectory data through the coordinated correction of inertial calculation and map geometric constraints, providing a reliable physical benchmark for visual perception consistency assessment.

[0094] Optionally, the method further includes: obtaining the vehicle's location information and the road geometry information of the road where the vehicle is currently located; and determining whether the vehicle is in a preset curve scenario based on the vehicle's location information and the road geometry information.

[0095] The vehicle location information mentioned above refers to the spatial coordinates of the vehicle in the global coordinate system output by a high-precision positioning system. It represents the vehicle's current absolute or relative geographical location and is the fundamental input for determining whether the vehicle has entered a specific road area. It possesses high spatiotemporal resolution and stability. The high-precision positioning system can be GNSS / Real-Time Kinematic (RTK) technology or a fusion positioning module.

[0096] The aforementioned road geometry information refers to the spatial features related to road structure pre-stored in high-precision maps. Road geometry information includes, but is not limited to, curvature variations, heading angles, slopes, lane line shapes, and key landmarks. It describes the topological and geometric constraints of the road and serves as prior knowledge for identifying high-risk driving scenarios. These key landmarks can include tunnel entrances / exits, curve start and end points, etc.

[0097] The aforementioned preset curve scenarios refer to specific driving situations predefined based on road geometry and vehicle motion. The criteria for determining preset curve scenarios are based on joint constraints of position and road geometry, and are typically associated with highly dynamic or sensor-failure environments to trigger enhanced fusion strategies. These preset curve scenarios can include sharp bends at tunnel exits, or scenarios where visual sensors fail due to blinding glare or sudden changes in field of view.

[0098] In one optional embodiment, the vehicle control system acquires the vehicle's current positioning information and the corresponding road geometric parameters in a high-precision map, including the curvature trend of the road centerline, the continuity of the heading angle, and the road segment type identifier. The vehicle control system compares the vehicle's position with specific geometric feature areas pre-stored in the map to determine whether the vehicle has entered a road segment with significant spatial variation characteristics. These characteristics manifest as abrupt changes in road curvature or continuous large-scale turning of the heading angle, accompanied by specific environmental contexts, such as the boundary of adjacent structures or linear transition zones. If the spatial correlation between the vehicle's position and the aforementioned road geometric features satisfies a preset topology matching condition, the vehicle control system determines that the vehicle is in a preset curve scenario, thereby triggering a subsequent perception strategy adjustment mechanism. The aforementioned topology matching condition refers to a discrimination criterion established in a multi-sensor fusion perception system to verify whether the spatial structural relationship between the visual perception results and the theoretical trajectory generated by the high-precision map / inertial navigation conforms to the topological constraints of the road network. In another optional embodiment, firstly, the vehicle control system identifies the vehicle's relative spatial position to the tunnel structure using a high-precision map. If the vehicle is in the vicinity of the tunnel exit structure, it determines that the vehicle is ready to enter the exit section. Secondly, the vehicle control system combines road geometry information from the high-precision map to extract the curvature change characteristics and heading change trends of the continuous road segment after the tunnel exit. If the aforementioned road segment exhibits a significant abrupt change in curvature or a rapid deflection of the heading angle within a very short distance after the exit, the vehicle control system determines that there is a sharp curve after the tunnel exit. Only when both conditions are met—the vehicle being in the vicinity of the tunnel exit and the road exhibiting sharp curve geometry after the exit—do the vehicle control system confirm that the vehicle is in the preset curve scenario of a sharp curve at the tunnel exit.

[0099] This application embodiment achieves accurate and proactive identification of high-risk scenarios through joint discrimination of location and road geometry, enabling the system to switch to an enhanced perception strategy in advance, thereby improving environmental adaptability and functional safety.

[0100] Optionally, determining whether the vehicle is in a preset curve scenario based on vehicle location information and road geometry information includes: determining the target distance between the vehicle's current position and the preset position based on map information; determining the road curvature at the aiming point in the road based on road geometry information; determining the target heading angle difference based on the first heading angle of the current position and the second heading angle of the target position in the road, wherein the target position is a position in the road at a preset distance from the current position; and determining whether the vehicle is in a preset curve scenario based on the target distance, road curvature, and target heading angle difference.

[0101] The aforementioned preset location refers to a geographic reference point that is pre-marked in a high-precision map and associated with a specific high-risk driving scenario, such as the boundary of a tunnel exit or the starting point of a curve. It is used as the reference coordinate for determining the vehicle's position. The spatial attributes of the preset location are associated with the geometric features of the road and serve as the positioning anchor point for scene recognition.

[0102] The target distance mentioned above refers to the straight-line distance between the vehicle's current position and the preset position. It is used to quantify how close the vehicle is to the critical scenario and is the primary spatial threshold condition for determining whether to enter the risk triggering area. It reflects the vehicle's position evolution state in the global coordinate system.

[0103] The aforementioned preview points refer to reference sampling points selected along the centerline of the road ahead of the vehicle's direction of travel, used to assess the future geometric characteristics of the road. The preview point positions are typically dynamically set based on vehicle speed and system response latency, used to anticipate upcoming curvature changes and support proactive scenario assessment.

[0104] The road curvature mentioned above refers to the instantaneous bending degree of the road centerline at the aiming point. Calculated from road geometry information, it reflects the degree of turning at the aiming point and is a core geometric parameter for identifying high-dynamic scenarios such as sharp bends and lane change transitions. A larger road curvature value indicates a larger steering angle required by the vehicle. A smaller road curvature value indicates a smaller steering angle required by the vehicle.

[0105] The aforementioned first heading angle refers to the angle between the vehicle's current driving direction and the reference axis of the global coordinate system. It is calculated from the vehicle's attitude information, represents the vehicle's real-time orientation, and serves as the basis for judging the degree of matching between the vehicle's movement direction and the road design intent.

[0106] The aforementioned second heading angle refers to the angle between the tangent direction of the road centerline at the target location and the reference axis of the global coordinate system. It is extracted from map information and represents the expected driving direction of the road, used to assess the heading consistency between the vehicle and the road path.

[0107] The aforementioned target heading angle difference refers to the deviation between the first heading angle and the second heading angle, reflecting the degree of alignment between the vehicle's current orientation and the expected direction of the road ahead. It is a key dynamic criterion for identifying whether the vehicle is about to enter a large-angle turning area. The larger the difference, the more severe the turning.

[0108] In one optional embodiment, the vehicle control system calculates the spatial distance between the vehicle's current position and the boundary points of a preset feature area based on road geometry data provided by a high-precision map. Simultaneously, the vehicle control system selects several pre-aiming points along a preset range ahead of the vehicle, extracting local road curvature features to assess the curvature intensity of the road alignment. Further, the vehicle control system obtains the heading angle of the vehicle's current heading angle and the heading angle of the corresponding point on the road centerline at a preset distance ahead, calculating the difference between the two to reflect the severity of road turning. By comprehensively considering the changing trends and coupling relationships of the target distance, road curvature, and heading angle difference, a joint judgment is made using preset logical rules. If the target distance is within the effective response range, the road curvature exhibits non-linear changes, and the heading angle difference exceeds a dynamic threshold, it is determined that the vehicle has entered a preset curve scenario, thereby activating subsequent perception fusion strategies.

[0109] The aforementioned pre-defined judgment rules refer to a set of structured logical conditions and decision criteria predefined in the system design phase based on physical laws, engineering experience, and scenario characteristics, used to comprehensively evaluate multi-dimensional input information and determine the scenario.

[0110] This application embodiment achieves accurate and forward-looking identification of preset curve scenarios through joint discrimination of multi-dimensional space and geometric features, improves the real-time performance and robustness of the system response, and provides a reliable trigger basis for intelligent perception switching.

[0111] Optionally, based on the target distance, road curvature, and target heading angle difference, determining whether the vehicle is in a preset curve scenario includes: determining that the vehicle is in a preset curve scenario in response to the target distance being less than a preset distance threshold, the road curvature being less than a preset curvature threshold, and the target heading angle difference being greater than a preset heading angle threshold; and determining that the vehicle is not in a preset curve scenario in response to the target distance being greater than or equal to a preset distance threshold, or the road curvature being greater than or equal to a preset curvature threshold, or the target heading angle difference being less than or equal to a preset heading angle threshold.

[0112] The aforementioned preset distance threshold refers to a spatial critical value set to identify specific driving scenarios. It is used to determine whether the vehicle has entered the potential influence range of the specific driving scenario. If the distance is below the aforementioned spatial critical value, it indicates that the vehicle has approached the critical area, triggering the scenario activation logic. The preset distance threshold is set based on a comprehensive safe distance considering both sensor response latency and system control requirements.

[0113] The aforementioned preset curvature threshold refers to the lower limit of curvature in road geometry used to distinguish between ordinary curves and high-risk sharp curves. A curvature below the preset threshold indicates a gentler road curvature. This preset curvature threshold is used to filter out low-risk curvature changes, ensuring that the fusion strategy is activated only in scenarios where enhanced perception is truly needed.

[0114] The aforementioned preset heading angle threshold refers to the maximum allowable deviation between the vehicle's current heading and the expected heading of the road ahead. A heading angle higher than the preset heading angle threshold indicates that the vehicle is about to face a drastic change in direction. It is the core dynamic criterion for identifying high-dynamic conditions such as sharp turns and large-angle curves, and is used to eliminate interference from gradual steering or straight-line driving.

[0115] In one optional embodiment, if the vehicle control system detects that the target distance is less than a preset distance threshold, the road curvature is less than a preset curvature threshold, and the target heading angle difference is greater than a preset heading angle threshold, then the vehicle is determined to be in a preset curve scenario. If the vehicle control system detects that the target distance is greater than a preset distance threshold, the road curvature is greater than a preset curvature threshold, and the target heading angle difference is less than a preset heading angle threshold, then the vehicle is determined not to be in a preset curve scenario.

[0116] This application embodiment achieves accurate and error-resistant identification of preset curve scenarios through triple threshold collaborative judgment, ensuring that the enhancement strategy is activated only at truly high-risk moments, thus balancing security and system efficiency.

[0117] Figure 2 This is a schematic diagram of tunnel exit curve scene recognition provided in an embodiment of this application, as shown below. Figure 2 As shown, the vehicle control system acquires the vehicle's real-time location information through the Global Navigation Satellite System and queries a high-precision map to obtain the tunnel exit coordinates and the geometric data of the road ahead. Based on the vehicle's current location and the tunnel exit coordinates, the distance to the tunnel exit is calculated and compared with a preset threshold condition. If the preset threshold condition is met, the system further calculates the curvature or heading angle change of the road ahead based on the road geometry of the high-precision map to perform a road geometry judgment condition analysis. Subsequently, based on the road geometry judgment condition, if the road geometry judgment condition is met, a high-risk scenario mode is triggered. If the road geometry judgment condition is not met, the system maintains the normal mode. If the preset threshold condition is not met, the system maintains the normal mode.

[0118] This application embodiment first performs preset curve scene recognition, which is the condition for triggering this application embodiment. Taking a sharp curve at a tunnel exit as an example, tunnel exit curve scene recognition needs to quickly and accurately determine whether a vehicle has entered that specific high-risk scene. Tunnel exit curve scene recognition includes position determination and road geometry determination.

[0119] This embodiment first performs a location determination, inputting the vehicle's latitude, longitude, and altitude coordinates provided by the GNSS / RTK positioning module. First, obtain a high-precision map. Second, query the high-precision map to get the coordinates of the nearest tunnel exit. The distance between the current position and the target position at the preset location is calculated using Euclidean distance as an example. Based on the preset distance threshold Make a judgment. The judgment condition is if... ,For example, If this condition is met, it means the vehicle is about to exit the tunnel.

[0120] When determining road geometry, first input the information provided by the high-precision map showing the front of the vehicle. The set of points representing the centerline of the road within the specified range. It can be 100 meters, and can be set according to needs in actual use. There are two methods for road geometry judgment, including Method 1 and Method 2. The two methods are logically complementary and together constitute a double verification for large corner scenarios.

[0121] Method 1 can determine road geometry using curvature. First, curve fitting is performed on the set of points along the road centerline. Second, the curvature at a point in front of the current vehicle's position is calculated. Road curvature The calculation formula is as follows:

[0122] ;

[0123] Method 2 can determine road geometry by adjusting the heading. First, calculate the road centerline at the vehicle's current position and the target position ahead. The difference in heading angle at the location .

[0124] For method 1, if the road curvature Preset curvature threshold ,For example Then it is a sharp bend. For method 2, if... Preset heading angle threshold This is a large turn, for example. .

[0125] The scene is triggered when a sharp turn or large angle is met. That is:

[0126] ;

[0127] Finally, output the scene flags. This indicates that the state is active.

[0128] For example, a vehicle is traveling inside a tunnel and is 30 meters away from the tunnel exit. The high-precision map shows that the tunnel exit is immediately followed by a left-turn lane with a radius of 150 meters (R=150m<200m). The system immediately triggers... Entering the high-risk scenario mode of the tunnel exit curve.

[0129] Figure 3 This is a schematic diagram of theoretical trajectory and feature generation based on dynamic consistency provided in an embodiment of this application, as shown below. Figure 3As shown, the vehicle control system first acquires raw data from the inertial measurement unit, including angular velocity and linear acceleration, and inputs it into the inertial navigation system for mechanical arrangement and trajectory calculation to generate the vehicle's initial motion state sequence. Subsequently, by combining high-precision map geometric information, the tight coupling filter of the inertial navigation system / map is constrained and improved to effectively suppress the accumulation error of inertial drift, thereby generating a high-precision theoretical trajectory. Based on this, theoretical features are extracted from the high-precision theoretical trajectory and the high-precision map, including the vehicle's lateral offset within the lane, the deviation between the heading angle and the road heading, and the curvature of the road ahead, as authoritative reference benchmarks for subsequent visual perception consistency assessment.

[0130] The high-risk scenario of entering a tunnel exit curve was identified, which involved generating a theoretical trajectory and features based on dynamic consistency. Upon scenario activation, the vehicle control system requires a vision-independent, highly reliable, and continuous reference benchmark. This benchmark is generated through tight coupling between the Inertial Navigation System (INS) and a high-precision map. The INS provides high-frequency, continuous attitude and velocity data, while the high-precision map provides geometric constraints. The combination of these two components generates a realistic and reliable vehicle trajectory, which can serve as a reference standard.

[0131] First, INS mechanics programming is performed, inputting raw IMU data, which consists of angular velocity ω and acceleration a. Then, by integrating ω and a using the inertial navigation differential equations, the vehicle's position, velocity, and attitude (PVT) are obtained.

[0132] ;

[0133] ;

[0134] ;

[0135] Output the trajectory calculated by INS Subsequently, INS / map tight coupling correction was performed, input... The coordinates of a high-precision map are used, along with road geometry such as lane centerlines, to constrain the cumulative error of the Inertial Navigation System (INS). For example, a Kalman filter is used to "pull" the INS-calculated position onto the road centerline of the map. The resulting high-precision theoretical trajectory is then corrected using the map. Unaffected by light or weather, it serves as the standard for subsequent comparisons.

[0136] In theoretical feature extraction, from Extract three key features from the trajectory and map. , and . This represents the vehicle's lateral offset within the lane. It is calculated by determining the vertical distance from the vehicle's position to the lane centerline. vehicle heading angle Deviation from road heading ,Right now . The curvature of the road ahead can be directly queried from a high-precision map.

[0137] For example, at the tunnel exit curve, the INS / map combination calculates the vehicle's theoretical trajectory as follows: the vehicle is traveling with a heading angle deviation of 2.5 degrees, a lateral deviation of 0.1 meters from the centerline, and the curvature of the road ahead is 0.0067 (1 / 150m).

[0138] Figure 4 This is a schematic diagram illustrating the calculation of visual perception feature extraction and dynamic consistency index according to an embodiment of this application, as shown below. Figure 4 As shown, the vehicle control system acquires visual information of the current environment through camera images and inputs it into a visual perception algorithm to identify key environmental features such as lane lines and road boundaries, thereby extracting visual features. Subsequently, the extracted visual features are compared item by item with theoretical features, and the difference is calculated to quantify the degree of deviation between the visual perception result and the high-precision theoretical trajectory. Finally, a weighted fusion is performed to obtain a comprehensive dynamic consistency index. The aforementioned differences include lateral position deviation, heading angle deviation, and curvature deviation, used to objectively evaluate the reliability of the visual sensor in the current scenario and provide a basis for subsequent dynamic weight adjustment.

[0139] When extracting visual perception features and calculating the consistency index of dynamics, the "world seen by vision" is compared with the "theoretical real world". If the two are highly consistent, it indicates that the vision is reliable; if the deviation is large, it indicates that the vision has been interfered with.

[0140] First, input a camera image. Then, using a trained deep learning model, such as LaneNet (a lane detection network), You Only Look Once (YOLO), or traditional image processing algorithms, identify lane lines in the image. Next, calculate... , and .

[0141] To calculate the distance from the vehicle to the left / right lane lines from the image, convert it to a lateral position in the world coordinate system. To fit the lane line, calculate the angle between the vehicle's heading and the lane direction. To extract curvature from the fitted lane line model.

[0142] Output visual perception features Then, the dynamic consistency index D is calculated, and the input is... ,Right now , and Further calculations , and .

[0143] ;

[0144] ;

[0145] ;

[0146] Comprehensive indicators ,in , , The D value is a weighting coefficient set according to the importance of each feature. The smaller the D value, the more consistent the visual perception is with the physical truth, and the more reliable the vision. The larger the D value, the greater the discrepancy between the visual perception and the physical truth, and the less reliable the vision.

[0147] For example, under normal circumstances, visual perception , , . and and Similarly, it is very small. A low overall D value (e.g., 0.05) indicates that vision is functioning normally.

[0148] Strong light at tunnel exit causes blindness: The strong light causes the visual algorithm to mistake shadows for lane lines. The distance suddenly jumped to 1.5 meters, making it appear as if the vehicle was about to veer off the road. It remains a stable 0.10 meters. The instantaneous increase to 1.4 meters caused the overall D value to rise sharply to 2.5, and the system immediately judged that visual perception was unreliable.

[0149] Figure 5 This is a schematic diagram of dynamic arbitration and fusion of multi-sensor confidence based on a consistency index, provided in an embodiment of this application. Figure 5 As shown, a weighted mapping function is constructed based on the dynamic consistency index. This function takes the comprehensive index value as input and dynamically generates the confidence weights of the visual sensor. Subsequently, the visual perception results, along with those from the inertial navigation system and map perception, are used as fusion inputs and weighted fusion is performed using the weighted mapping function. Finally, a robust perception result is output, effectively improving the overall perception robustness and safety of the system in high-risk scenarios.

[0150] In the dynamic arbitration and fusion of multi-sensor confidence based on consistency indices, the decision to trust visual data or INS / map data is dynamically made based on the magnitude of the comprehensive index D. A large D value indicates that visual data is unreliable, and trajectory planning relies more on INS / map data.

[0151] Weight mapping function The core formula for the design is as follows:

[0152] ;

[0153] The above weight mapping function It is an sigmoid function. Wherein, It is an inflection point, if , k is the slope, controlling the speed at which the weight decreases from 1 to 0. If D is very small, A value close to 1 indicates that visual data can be trusted. If D is large, W_vision approaches 0, indicating that INS / maps can be trusted. In the intermediate transition zone, the weights change smoothly.

[0154] Ultimately merged into .in, It refers to the perception results provided by the INS / map combination, such as vehicle pose.

[0155] For example, at the beginning of the scene Calculated according to the function The fusion results are 95% derived from vision, and the system operates smoothly. Upon exiting the tunnel, the intense light causes blindness, and D instantly jumps to 2.5. The calculated... The fusion result immediately shifts to 95% from INS / map. The vehicle won't swerve due to a sudden visual "misreading," but will continue driving along the stable trajectory calculated by INS / map. Once out of the bright light zone, vision recovers, and the D-value gradually decreases. As the temperature gradually rises, the system smoothly switches back to vision-driven mode.

[0156] This application also provides an apparatus 60 for a vehicle control method. Figure 6 This is a schematic diagram of a vehicle control device provided in an embodiment of this application, as shown below. Figure 6 As shown, it includes: an acquisition module 602, an extraction module 604, an evaluation module 606, and a control module 608.

[0157] The acquisition module 602 is used to acquire vehicle trajectory data and visual perception images in response to detecting that the vehicle is in a preset curve scenario; the extraction module 604 is used to extract features from the vehicle trajectory data and visual perception images based on multiple dynamic indicators to obtain vehicle state features and visual perception features corresponding to the multiple dynamic indicators, wherein different dynamic indicators are used to represent state quantities of different path features of the vehicle traveling on the road; the evaluation module 606 is used to perform consistency evaluation on the vehicle state features and visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and visual perception features on multiple dynamic indicators; the control module 608 is used to perform weighted fusion of the vehicle trajectory data and visual perception images based on the target evaluation value to obtain a target fusion result, and control the vehicle based on the target fusion result.

[0158] The control module is also used to determine the visual weight value corresponding to the visual perception image based on the preset correspondence and target evaluation value; to determine the state weight value corresponding to the vehicle trajectory data based on the difference between the preset value and the visual weight value; and to perform weighted fusion of the visual perception image and the vehicle trajectory data based on the visual weight value and the state weight value to obtain the target fusion result.

[0159] The evaluation module is also used to determine the feature difference values ​​corresponding to multiple dynamic indicators based on vehicle state characteristics and visual perception characteristics; and to perform weighted fusion of the feature difference values ​​corresponding to multiple dynamic indicators based on the weight values ​​of multiple dynamic indicators to obtain the target evaluation value.

[0160] The device is also used to acquire angular velocity and acceleration collected by the inertial measurement unit in the vehicle; integrate the angular velocity and acceleration to obtain the vehicle's position information, vehicle speed information and vehicle attitude information; and predict the vehicle's trajectory based on the vehicle position information, vehicle speed information and vehicle attitude information to obtain vehicle trajectory data.

[0161] The device is also used to acquire map information of the road where the vehicle is currently located; based on the vehicle's position information, speed information, and attitude information, it predicts the vehicle's trajectory to obtain initial vehicle trajectory data; it constrains the position information in the initial vehicle trajectory data to the road centerline corresponding to the map information, and constrains the attitude information in the initial vehicle trajectory data to the tangent direction of the road centerline to obtain vehicle trajectory data.

[0162] The device is also used to acquire the vehicle's location information and the road geometry information of the road where the vehicle is currently located; based on the vehicle's location information and the road geometry information, it determines whether the vehicle is in a preset curve scenario.

[0163] The device is also used to determine the target distance between the vehicle's current position and a preset position based on map information; to determine the road curvature at the pre-aiming point based on road geometry information; to determine the target heading angle difference based on the first heading angle of the current position and the second heading angle of the target position on the road, wherein the target position is a position on the road at a preset distance from the current position; and to determine whether the vehicle is in a preset curve scenario based on the target distance, road curvature, and target heading angle difference.

[0164] The device is also used to determine that the vehicle is in a preset curve scenario when the target distance is less than a preset distance threshold, the road curvature is less than a preset curvature threshold, and the target heading angle difference is greater than a preset heading angle threshold; and to determine that the vehicle is not in a preset curve scenario when the target distance is greater than or equal to a preset distance threshold, or the road curvature is greater than or equal to a preset curvature threshold, or the target heading angle difference is less than or equal to a preset heading angle threshold.

[0165] This application also provides an electronic device 70. Figure 7 This is a structural diagram of an electronic device provided in an embodiment of this application, as shown below. Figure 7 As shown, it includes a memory 710 and a processor 720, wherein the memory 710 is used to store computer programs; and the processor 720 is used to execute the programs stored in the memory 710 to implement the vehicle control method described in any embodiment of this application.

[0166] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the vehicle control method described in any embodiment of this application.

[0167] In this application, "multiple" refers to two or more.

[0168] In this application, unless otherwise expressly defined, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0169] The terms “first,” “second,” “third,” “fourth,” etc., in this application (if any) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0170] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0171] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if the method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if the method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.

[0172] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle control method, characterized in that, include: In response to detecting that the vehicle is in a preset curve scenario, the vehicle trajectory data and visual perception image of the vehicle are acquired; Based on multiple dynamic indicators, feature extraction is performed on the vehicle trajectory data and the visual perception image to obtain the vehicle state features and visual perception features corresponding to the multiple dynamic indicators respectively. Among them, different dynamic indicators are used to represent the state quantities of different path features of the vehicle traveling on the road. A consistency evaluation is performed on the vehicle state features and the visual perception features to obtain a target evaluation value, wherein the target evaluation value is used to represent the degree of difference between the vehicle state features and the visual perception features on the multiple dynamic indicators; The vehicle trajectory data and the visual perception image are weighted and fused based on the target evaluation value to obtain the target fusion result, and the vehicle is controlled based on the target fusion result.

2. The method according to claim 1, characterized in that, The vehicle trajectory data and the visual perception image are weighted and fused based on the target evaluation value to obtain the target fusion result, including: Based on the preset correspondence and the target evaluation value, the visual weight value corresponding to the visual perception image is determined; Based on the difference between the preset value and the visual weight value, the state weight value corresponding to the vehicle trajectory data is determined. Based on the visual weight value and the state weight value, the visual perception image and the vehicle trajectory data are weighted and fused to obtain the target fusion result.

3. The method according to claim 1, characterized in that, Based on the weight values ​​of the multiple dynamic indicators, a consistency evaluation is performed on the vehicle state features and the visual perception features to obtain a target evaluation value, including: Based on the vehicle state characteristics and the visual perception characteristics, the feature difference values ​​corresponding to the multiple dynamic indicators are determined; The target evaluation value is obtained by weighting and fusing the feature difference values ​​corresponding to the multiple dynamic indicators based on their weight values.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: The angular velocity and acceleration collected by the inertial measurement unit in the vehicle are obtained; Integrating the angular velocity and the acceleration yields the vehicle's position information, vehicle velocity information, and vehicle attitude information. Based on the vehicle location information, vehicle speed information, and vehicle attitude information, the vehicle trajectory is predicted to obtain the vehicle trajectory data.

5. The method according to claim 4, characterized in that, Based on the vehicle location information, the vehicle speed information, and the vehicle attitude information, the vehicle trajectory data is determined, including: Obtain map information of the road where the vehicle is currently located; Based on the vehicle location information, the vehicle speed information, and the vehicle attitude information, the vehicle trajectory is predicted to obtain initial vehicle trajectory data; The position information in the initial vehicle trajectory data is constrained to the road centerline corresponding to the map information, and the attitude information in the initial vehicle trajectory data is constrained to the tangent direction of the road centerline to obtain the vehicle trajectory data.

6. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain the vehicle's location information and the road geometry information of the road where the vehicle is currently located; Based on the vehicle location information and the road geometry information, it is determined whether the vehicle is in the preset curve scenario.

7. The method according to claim 6, characterized in that, Determining whether the vehicle is in the preset curve scenario based on the vehicle location information and the road geometry information includes: Based on map information, determine the target distance between the vehicle's current location and the preset location; Based on the road geometry information, the road curvature at the pre-aiming point in the road is determined; Based on the first heading angle of the current position and the second heading angle of the target position in the road, a target heading angle difference is determined, wherein the target position is a position in the road that is at a preset distance from the current position; Based on the target distance, the road curvature, and the target heading angle difference, it is determined whether the vehicle is in the preset curve scenario.

8. The method according to claim 7, characterized in that, Based on the target distance, the road curvature, and the target heading angle difference, determining whether the vehicle is in the preset curve scenario includes: In response to the target distance being less than a preset distance threshold, the road curvature being less than a preset curvature threshold, and the target heading angle difference being greater than a preset heading angle threshold, it is determined that the vehicle is in the preset curve scenario; In response to the target distance being greater than or equal to the preset distance threshold, or the road curvature being greater than or equal to the preset curvature threshold, or the target heading angle difference being less than or equal to the preset heading angle threshold, it is determined that the vehicle is not in the preset curve scenario.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing a program stored in memory to implement the method described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 8.

Images