Methods and systems for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles

Abstract

This invention relates to the field of electric vehicle control technology, and provides a method and system for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles. The method includes: acquiring vehicle driving information, preceding vehicle point cloud, and image data using onboard radar and cameras based on a unified coordinate system to ensure spatial alignment of perception data; selecting the closest point to the vehicle from the preceding vehicle point cloud as a benchmark and constructing a horizontal plane to simplify the three-dimensional point cloud into two-dimensional data, eliminating redundant information and reducing computational load; classifying the target point cloud into three categories according to the spatial distribution direction to accurately match the relative positions of the preceding vehicle and the vehicle; adapting and fitting different point cloud sets to generate preceding vehicle boundaries, i.e., warning lines; detecting the position and deflection angle of the warning lines in real time, and determining the preceding vehicle's lane-changing intention by combining the included angle, turn signal, and offset; and controlling the vehicle to decelerate or brake according to the preceding vehicle category, thereby improving the decision-making response speed and driving safety of intelligent driving in complex road conditions.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle control technology, and more specifically, to a method and system for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] Intelligent assisted driving is a core technology for electric vehicle control. Accurate identification and safe decision-making regarding the driving intentions of surrounding vehicles in complex traffic scenarios directly determine the safety and efficiency of vehicle operation during customer commutes. Currently, most vehicle perception systems utilize multi-sensor fusion solutions, such as vehicle radar and cameras. By collecting environmental point cloud data from radar and acquiring visual image data from cameras, comprehensive perception of the position, motion, and driving behavior of vehicles ahead is achieved, providing underlying perception support for active obstacle avoidance, speed control, and path planning.

[0004] However, existing driver intent recognition technologies still have significant shortcomings: conventional multi-sensor fusion solutions often process the entire raw data directly, resulting in high data redundancy and high computational consumption. In complex road conditions such as dense intersections of multiple vehicles, sudden lane changes, and lane congestion, problems such as data processing lag and slow decision-making response easily arise, making it difficult to meet the dual requirements of real-time performance and safety for electric vehicles. Even with perception solutions that integrate LiDAR and communication, the core problem of low point cloud data processing efficiency remains unresolved.

[0005] For example, Chinese patent CN202311538770.7 discloses a method for vehicle driving intention recognition based on LiDAR and visible light communication. While this patent achieves basic intention recognition through the fusion of LiDAR point cloud detection and visible light communication data, the solution consistently uses all point cloud data for vehicle detection, tracking, and state calculation without filtering or simplifying redundant point clouds. In complex scenarios with dense traffic and changing road conditions, full point cloud computing leads to unnecessary computational waste and is susceptible to interference from point clouds, resulting in a significant decrease in the real-time performance and accuracy of intention recognition. This makes it unsuitable for the practical application requirements of electric vehicles in complex road conditions for rapid and accurate recognition of the driving intentions of vehicles ahead.

[0006] To address the issues of redundancy, low efficiency, and insufficient accuracy in existing technologies under complex road conditions, there is an urgent need for a method and system for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles. This system should optimize the perception data processing flow, improve the reliability and response speed of intention recognition in complex scenarios, and meet the practical requirements of intelligent driving of electric vehicles. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles, so as to improve the above-mentioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: In a first aspect, this application provides a method for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles, including an onboard radar and an onboard camera mounted on the electric vehicle and operating in a unified coordinate system. The method includes: Obtain the vehicle's driving direction and current position; acquire valid point cloud data of the vehicle in front using the vehicle-mounted radar; acquire image data of the vehicle in front using the vehicle-mounted camera; From the point cloud data of each preceding vehicle, select the point closest to the current vehicle as the reference point for the corresponding preceding vehicle, and establish the horizontal plane of the preceding vehicle in the coordinate system based on the reference point; select the point cloud data of the preceding vehicle located on the horizontal plane as the target point cloud set. When the target points are clustered along two directions and the two directions are within a preset angle range, it is identified as the first cluster of points; when the target points are clustered along the direction of travel of the vehicle, it is identified as the second cluster of points; when the target points are clustered along the lateral direction of the vehicle, it is identified as the third cluster of points. The first point cloud is subjected to piecewise straight line fitting, and the second and third point clouds are subjected to straight line fitting. The straight line fitting includes: in the corresponding target point cloud, a preset number of target position points located on the outermost side of the vehicle in front are selected, and the target position points are fitted into the reference straight line of the vehicle in front; the points located at the end positions in the corresponding target point cloud are mapped to the reference straight line as endpoints; and the warning line of the vehicle in front is generated based on the reference straight line with the endpoints as constraints. The position and deflection angle of each warning line are detected in real time. If the vehicle in front of the first point cluster meets the preset lane change rules, the vehicle controller of this vehicle will control the vehicle to decelerate or brake according to the distance. If the vehicle in front of the second point cluster meets the preset lane change rules, the vehicle controller will brake in an emergency. If the distance between the warning line of the vehicle in front of the third point cluster and this vehicle is less than the first preset distance, the vehicle controller will decelerate or brake. The preset lane-changing rules include: the angle between the warning line of the preceding vehicle and the direction of travel of the current vehicle is greater than a preset angle, or the turn signal of the preceding vehicle towards the direction of the current vehicle's lane is continuously illuminated in the image data within a preset time, or the warning line shifts towards the direction of the current vehicle's lane by a second preset distance within a preset time.

[0008] Furthermore, the step of selecting the point closest to the current vehicle as the reference point for the corresponding preceding vehicle specifically includes: Based on the time-of-flight ranging principle of vehicle-mounted radar, multiple echo points with the shortest time of reflection are obtained from all point cloud data of the preceding vehicle. Multiple echo points are clustered according to their spatial distribution density, and the point located at the center of the clustered area is extracted as the reference point.

[0009] Furthermore, the steps for establishing the horizontal plane of the preceding vehicle in the coordinate system using a reference point specifically include: Centered on the reference point, generate a virtual plane in a unified coordinate system whose normal vector is parallel to the direction of the vehicle's gravity; define the virtual plane as the horizontal plane of the vehicle in front.

[0010] Furthermore, the step of identifying a target point cluster as the first point cluster when the target points are distributed along two directions and the two directions are within a preset angle range specifically includes: Under a unified coordinate system, the spatial distribution of the acquired target point cloud is traversed across the entire domain. The changes in point cloud distribution density of adjacent spatial positions within the target point cloud are statistically analyzed segment by segment. Spatial directions in which the density changes continuously from large to small are identified. Each complete and uninterrupted continuous spatial direction is determined as a distribution direction of the target point cloud. When two sets of independent distribution directions are identified, the two directions are fitted with a linear function to obtain two corresponding objective functions. The distribution angle between the two objective functions is calculated. When the distribution angle is within the preset angle range, the target point set is identified as the first point set.

[0011] Furthermore, before performing a global spatial distribution traversal on the acquired target point cloud, the following steps are also included: The vertical coordinate information of the target point cloud is deleted, and only the two-dimensional coordinates on the horizontal plane are used for subsequent direction identification and angle calculation.

[0012] Furthermore, the step of statistically analyzing the changes in point cloud distribution density at adjacent spatial locations within the target point cloud segment by segment specifically includes: The area covered by the cluster of target points on the horizontal plane is divided into multiple continuous grids by equal intervals. The number of point clouds contained in each grid is counted sequentially, and the difference in the number of point clouds in adjacent spatial grids is compared to obtain the change in point cloud distribution density at adjacent spatial locations within the target point cloud set.

[0013] Furthermore, the step of selecting a predetermined number of target location points located on the outermost edge of the preceding vehicle specifically includes: In the second point cloud, calculate the distance from each point in the vehicle's driving direction to the origin of the coordinate system, and select the preset number of points with the smallest distance as the outermost target position points; In the third point cluster, calculate the distance from each point on the vehicle's lateral side to the origin of the coordinate system, and select the preset number of points with the smallest distance as the outermost target position points.

[0014] Furthermore, it also includes: When the target point cloud is distributed along three or more directions, it is identified as the fourth point cloud; the fourth point cloud is fitted with a multi-segment continuous straight line; the multi-segment continuous straight line fitting includes: fitting all points in the fourth point cloud to form a piecewise linear function; In the piecewise linear function, sub-functions whose slope and the lateral slope error of the vehicle are within a preset range are selected. The distance between the two farthest sub-functions in the driving direction is calculated. If the distance is greater than the threshold, the vehicle is judged as dangerous. The line segment containing the sub-function closest to the vehicle in the direction of travel is used as the warning line. The vehicle is controlled to maintain a third preset distance from the warning line in the direction of travel, or the vehicle is controlled to accelerate and exceed the vehicle in front corresponding to the fourth point cluster.

[0015] Furthermore, it also includes: The height of the preceding vehicle is calculated based on the point cloud data of each preceding vehicle; For the vehicle in front corresponding to the first point cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side and the wide side of the vehicle in front. Based on the height of the vehicle in front and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front of the second cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side of the vehicle in front. Based on the height of the vehicle in front, the preset assumed width and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front corresponding to the third point cluster, the length of the warning line of the vehicle in front is used as the constraint of the width of the vehicle in front. Based on the height of the vehicle in front, the preset assumed length and the preset vehicle model, a real-time model of the vehicle in front is established; the real-time model is displayed on the vehicle screen.

[0016] Secondly, this application also provides a complex road condition intention recognition and decision-making system for intelligent driving of electric vehicles, including: The perception and acquisition module is used to obtain the vehicle's driving direction and current position; to obtain effective point cloud data of the vehicle in front through the vehicle-mounted radar; and to obtain image data of the vehicle in front through the vehicle-mounted camera. The point cloud filtering module is used to filter the points closest to the current vehicle from the point cloud data of each preceding vehicle as the reference point of the corresponding preceding vehicle, and to establish the horizontal plane of the preceding vehicle in the coordinate system based on the reference point; and to filter the point cloud data of the preceding vehicle located on the horizontal plane as the target point cloud set. The point cloud classification module is used to identify a target point cloud as a first point cloud if the target point cloud is distributed along two directions and the two directions are within a preset angle range; to identify a target point cloud as a second point cloud if the target point cloud is distributed along the vehicle's driving direction; and to identify a target point cloud as a third point cloud if the target point cloud is distributed laterally along the vehicle. The fitting module is used to perform piecewise straight line fitting on the first point cloud and straight line fitting on the second and third point clouds. The straight line fitting includes: in the corresponding target point cloud, selecting a preset number of target position points located on the outermost side of the vehicle in front, and fitting the target position points into a reference straight line of the vehicle in front; mapping the points located at the end positions in the corresponding target point cloud to the reference straight line as endpoints; and generating a warning line for the vehicle in front based on the reference straight line with the endpoints as constraints. The decision module is used to detect the position and deflection angle of each warning line in real time. If the vehicle in front of the first point cluster meets the preset lane change rules, the vehicle controller of this vehicle will control the vehicle to decelerate or brake according to the distance. If the vehicle in front of the second point cluster meets the preset lane change rules, the vehicle controller will brake in an emergency. If the distance between the warning line of the vehicle in front of the third point cluster and this vehicle is less than the first preset distance, the vehicle controller will decelerate or brake. The preset lane-changing rules include: the angle between the warning line of the preceding vehicle and the direction of travel of the current vehicle is greater than a preset angle, or the turn signal in the direction of the current vehicle's lane is continuously illuminated in the image data within a preset time, or the warning line is shifted to the direction of the current vehicle's lane by a second preset distance within a preset time.

[0017] The beneficial effects of this invention are as follows: This invention relies on a unified coordinate system for vehicle-mounted radar and cameras to achieve spatial alignment of multi-source perception data. By acquiring the vehicle's driving position, the point cloud of the preceding vehicle, and image data, it ensures accurate and offset perception information. Then, it selects the closest point to the vehicle from the preceding vehicle's point cloud as a reference point and constructs a horizontal plane. Target point clouds within this horizontal plane are then filtered, simplifying the three-dimensional point cloud into two-dimensional data. This effectively eliminates redundant information and reduces computational load. Simultaneously, based on the spatial distribution direction of the target point cloud, it classifies them into three categories, accurately matching the relative position and attitude of the preceding vehicle and the vehicle itself, providing a basis for subsequent differentiated processing. Then, different points... Yunji employs an adaptive straight-line fitting method. By selecting the outermost point of the preceding vehicle to fit a reference straight line, mapping end points to constrain endpoints, and generating a warning line, it accurately outlines the body contour of the preceding vehicle to form its boundary. Finally, it detects the position and deflection angle of the warning line in real time. Combining the warning line angle, turn signal status, and warning line offset, it determines the preceding vehicle's lane-changing intention. It then controls the vehicle's deceleration or braking according to the preceding vehicle's category, adopting differentiated braking strategies for different risk levels. This allows for rapid identification of the preceding vehicle's driving intention and precise execution of hazard avoidance decisions, improving the decision-making response speed and driving safety of intelligent driving in complex road conditions. Attached Figure Description

[0018] Figure 1 A flowchart of a complex road condition intent recognition and decision-making method for intelligent driving of electric vehicles provided by the present invention; Figure 2 This is a schematic diagram showing the positions and radar range of the preceding vehicle and the vehicle itself in this invention; Figure 3 This is a schematic diagram of the point closest to the vehicle in the straight-line driving of the preceding vehicle e in this invention; Figure 4 This is a schematic diagram of the point closest to the vehicle in the lane change of the preceding vehicle e in this invention; Figure 5 This is a schematic diagram of the planar point cloud of the front vehicle e in this invention; Figure 6 This is a schematic diagram of the planar point cloud of the front vehicle g in this invention; Figure 7 This is a schematic diagram of the planar point cloud of the front vehicle f in this invention; Figure 8 This is a schematic diagram showing the positions of the truck and the vehicle itself, as well as the radar range, in this invention; Figure 9 This is a schematic diagram of the piecewise linear function corresponding to the truck in this invention; Figure 10 A schematic diagram of a complex road condition intention recognition and decision-making system for intelligent driving of electric vehicles provided for the invention.

[0019] In the diagram: 1. This vehicle's lane; 2. Adjacent lane; 3. Vehicle radar; 4. This vehicle; 5. Nearest point; 6. Truck; 7. Sub-function; 8. Nearest sub-function; 201. Perception and acquisition module; 202. Point cloud filtering module; 203. Point cloud classification module; 204. Fitting module; 205. Decision module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Example 1 like Figure 1 As shown in the embodiment of the present invention, a method for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles includes an onboard radar 3 and an onboard camera mounted on the electric vehicle in a unified coordinate system. The method includes: S101: Obtain the driving direction and current position of this vehicle 4; obtain valid point cloud data of the vehicle in front through the vehicle radar 3; obtain image data of the vehicle in front through the vehicle camera.

[0022] Specifically, relying on the vehicle-mounted inertial measurement unit, wheel speed sensor, and lane line recognition module, local parameter calculations are completed within a unified coordinate system. By collecting vehicle heading angle, wheel speed, and lane line visual feature signals, the current position and driving direction of vehicle 4 are determined, i.e., the specific lane (such as the left lane, middle lane, or right lane) and driving direction in the current road. Then, the vehicle-mounted radar 3, which is installed in the electric vehicle and uses the same coordinate system as the vehicle-mounted camera, acquires effective point cloud data of the preceding vehicle. Based on the time-of-flight ranging principle, the vehicle-mounted radar 3 emits detection signals in the driving direction of vehicle 4, receives the echo signals reflected by the preceding vehicle, and performs preprocessing such as clutter filtering and interference removal to select effective point cloud data that corresponds only to the preceding vehicle. The unified coordinate system can eliminate the spatial offset between the radar and the camera data, ensuring the spatial alignment of multi-source perception data. Finally, image data of the vehicle in front is acquired through the vehicle-mounted camera in the same unified coordinate system. The camera collects visual feature information such as the outline and turn signals of the vehicle in front, and completes basic preprocessing such as noise reduction and outline enhancement. This forms a complementary perception for the identification and locking of the vehicle in front with the three-point cloud data of the vehicle-mounted radar, making up for the perception blind spots of a single sensor. This provides comprehensive and accurate raw data support for subsequent selection, distribution identification and intent judgment of the vehicle in front, and improves the reliability of the target identification of the vehicle in front in small scenarios.

[0023] S102, from the point cloud data of each preceding vehicle, select the point 5 closest to the current vehicle 4 as the reference point of the corresponding preceding vehicle, and establish the horizontal plane of the preceding vehicle in the coordinate system with the reference point; select the point cloud data on the preceding vehicle located on the horizontal plane as the target point cloud set.

[0024] Specifically, point 5, the closest point to vehicle 4, is selected as the reference point for the preceding vehicle. The core principle is based on the time-of-flight ranging characteristics of the vehicle-mounted radar 3. The shorter the echo time, the closer the corresponding point cloud is to vehicle 4. This point most directly reflects the actual distance relationship between the preceding vehicle and vehicle 4, providing a precise reference for subsequent horizontal plane establishment and avoiding errors in subsequent point cloud selection and direction recognition due to reference point deviation. The formula corresponding to the time-of-flight ranging of the vehicle-mounted radar 3 is: (1) In the formula, The distance from the point cloud corresponding to a single echo to vehicle 4; The speed of electromagnetic wave propagation; This refers to the flight time of the radar detection signal.

[0025] Crucially, the nearest point, point 5, serves as the reference point for the virtual horizontal plane, not the vehicle's reference point. Therefore, as... Figure 2As shown in the diagram, e, f, and g represent vehicles in different states, and the dashed line at vehicle radar 3 indicates its emission range. Since vehicle g is located in lane 1, tracking it primarily involves the distance between the nearest point and vehicle 4. Therefore, even if the nearest point 5 changes due to attitude changes, it still meets the requirements for collision calculation. Thus, the nearest point 5 is still used to establish the horizontal plane for vehicle g. Vehicle f is parallel to vehicle 4. Although its state differs from vehicle g, the core principle for both is still their distance from vehicle 4. Even if the nearest point 5 on vehicle f changes, the calculation still uses its nearest point 5 as a reference to avoid collision.

[0026] Finally, for the vehicle e in adjacent lane 2, since there is a certain distance between them, to avoid misjudging intent due to changes in the closest point, it is necessary to find a point that does not change much as a reference. At this time, since the rear of the vehicle e is completely within the range of the vehicle radar 3, and according to geometric calculations, regardless of whether the vehicle e is traveling in a straight line (e.g., ... Figure 3 (as shown) or the lane change process (such as) Figure 4 As shown in the diagram, as long as the vehicle in front has not entered lane 1 of this vehicle, the area closest to this vehicle 4 remains unchanged. This closest area necessarily includes point 5, which is closest to this vehicle 4. Selecting this point as a benchmark can effectively avoid a lot of redundant calculations, realize a rapid judgment of the status of the vehicle in front, and thus adapt to the real-time decision-making needs under complex road conditions.

[0027] Specifically, based on the time-of-flight ranging principle of the vehicle-mounted radar 3, multiple echo points with the shortest flight times are obtained from the total point cloud data of the preceding vehicle. Since radar detection is subject to slight clutter interference, a single shortest flight time echo point may be inaccurate. Therefore, multiple echo points are clustered according to their spatial distribution density. A density clustering algorithm (such as DBSCAN) is used to group echo points with similar spatial locations into one category. The point located at the center of the clustered area is extracted as the reference point. The coordinates of the reference point are calculated by taking the arithmetic mean of the coordinates of multiple shortest flight time echo points, as follows: (2) In the formula, The coordinates of the reference point for the preceding vehicle; The number of echo points acquired by the radar in the shortest time; For the first The three-dimensional coordinates of the shortest echo point.

[0028] The core purpose of establishing a horizontal plane for the preceding vehicle in a coordinate system using a reference point is to reduce the amount of data computation and solve the problem in existing technologies where vehicle-mounted radar 3 needs to process a large amount of 3D point cloud data to extract reference features of the preceding vehicle, resulting in a large computational load and decision response that cannot be applied to complex road conditions. This invention simplifies 3D point cloud data into 2D data by constructing a horizontal plane, greatly reducing computational complexity. The principle is based on the spatial uniformity of a unified coordinate system, combined with the stability of the vehicle's 4 gravity direction, to construct a virtual plane parallel to the actual plane of the preceding vehicle. This plane constrains the 3D point cloud data of the preceding vehicle to the same 2D plane, eliminating redundant coordinate data in the vertical direction of 3D space and simplifying the data dimension. Specifically, with the reference point as the center, a virtual plane with a normal vector parallel to the vehicle's 4 gravity direction is generated in a unified coordinate system. The vehicle's 4 gravity direction can be determined by the gravity acceleration signal collected in real time by the vehicle-mounted inertial measurement unit, and its normal vector direction is consistent with the gravity acceleration direction. This virtual plane is then defined as the horizontal plane of the preceding vehicle. For example, if the z-axis is vertical in a unified coordinate system (opposite to the direction of gravity), then the normal vector of the horizontal plane is (0,0,1). With the reference point (x0,y0,z0) as the center, a plane with z=z0 is constructed as the horizontal plane of the vehicle in front. This plane can constrain all the point clouds of the vehicle body in the two-dimensional plane with z=z0, eliminating the coordinate differences of each point cloud in the z-axis direction, directly converting the three-dimensional point cloud into two-dimensional data, greatly reducing the amount of computation in subsequent data processing, and accurately corresponding to the horizontal area where the vehicle body is located. It can also simultaneously remove irrelevant point clouds such as ground reflections and air clutter, taking into account both data simplification and effectiveness.

[0029] Selecting point cloud data located on the horizontal plane on the preceding vehicle as the target point cloud set is a further implementation of the aforementioned simplification of 3D to 2D data conversion. Its core is to retain 2D point cloud data within the horizontal plane and eliminate redundant 3D data outside the plane, continuously reducing the computational load. The principle is to compare the consistency of the coordinates of each point cloud with the coordinates of the horizontal plane, retaining those that meet the requirements and eliminating invalid point clouds that deviate from the plane. Simultaneously, the 2D data dimension is solidified to avoid introducing redundant calculations of 3D coordinates in subsequent processing. For example, a coordinate deviation threshold of ±5cm is set, and point clouds whose z-coordinate difference from the horizontal plane z0 falls within this threshold are retained, forming the target point cloud set. This target point cloud set already belongs to 2D planar data, eliminating the need to process the vertical coordinates in 3D space.

[0030] S103, when the target points are clustered along two directions and the two directions are within a preset angle range, it is identified as the first cluster of points; when the target points are clustered along the driving direction of the vehicle 4, it is identified as the second cluster of points; when the target points are clustered along the lateral direction of the vehicle 4, it is identified as the third cluster of points. Specifically, this step, based on the spatial distribution characteristics of the target point cloud, corresponds to the relative lane position and attitude of the preceding vehicle and the current vehicle 4. By distinguishing the direction of the point cloud distribution, it achieves accurate classification of the preceding vehicle's position scene, providing targeted data support for subsequent straight-line fitting, warning line generation, and driving decision-making. Simultaneously, accurate classification simplifies subsequent data processing procedures and improves decision-making response efficiency. It should be noted that, as... Figure 2 As shown, Figure 2 This is a top-down view diagram, intended to show the range of the vehicle-mounted radar 3 on a horizontal plane. Figure 2 It can be clearly seen that the range that vehicle radar 3 can detect on the vehicle in front is [missing information]. Figure 2 The bolded line segment on the vehicle in front. The target point cloud data of the vehicle in different states is represented as a point cloud map (e.g., Figure 4 , Figure 5 , Figure 6 After displaying the data in the form shown (as shown), it was found that its distribution shape is basically consistent with the bolded line segment in Figure 2. Therefore, based on this principle, step S103 is performed.

[0031] When the target points are clustered and distributed in two directions, such as Figure 5 As shown, if the two directions are within a preset angle range, it is considered the first point cluster; that is, at this time, all the corresponding vehicles in front are located in front of vehicle 4, and these vehicles are located in adjacent lane 2. The principle is that there is a certain angle between the vehicle body of the vehicle in front of adjacent lane 2 and the driving direction of vehicle 4, and its vehicle body point cloud will show two continuous distribution directions with an angle close to perpendicular on the horizontal plane. Through this distribution feature, the vehicles in front of adjacent lane 2 can be accurately distinguished. When the target point cluster is distributed along the driving direction of vehicle 4, such as... Figure 7 As shown, this is identified as the second point cloud cluster; that is, the corresponding preceding vehicle is partially located in front of vehicle 4, and the preceding vehicle is located in adjacent lane 2, which is equivalent to being next to vehicle 4. The principle is that the preceding vehicle in adjacent lane 2, which is parallel or closely following, has a body extension direction that is basically consistent with the driving direction of vehicle 4. The point cloud will be concentrated along the driving direction of vehicle 4, which can quickly identify the preceding vehicle in parallel adjacent lane 2. When the target point cloud cluster is distributed laterally along vehicle 4, such as Figure 6 As shown, it is identified as the third point cloud, which means that the vehicle in front and the vehicle 4 are in the same lane and directly in front of the vehicle 4. The principle is that the width of the vehicle in front in the same lane is consistent with the lateral direction of the vehicle 4. The point cloud will be distributed along the lateral direction of the vehicle 4, which can directly locate the vehicle in front in the same lane and provide accurate basis for emergency avoidance decision-making.

[0032] The step of identifying a target point cloud as the first point cloud when the target point cloud is distributed along two directions and the two directions are within a preset angle range includes: firstly deleting the vertical coordinate information in the target point cloud, and only using the two-dimensional coordinates on the horizontal plane to participate in subsequent direction identification and angle calculation. The principle is that S102 has constrained the point cloud within the horizontal plane, and the vertical coordinates are redundant data. After deletion, vertical interference can be eliminated, and the distribution characteristics of the point cloud on the horizontal plane can be focused, thereby reducing the amount of data calculation and improving the vehicle response speed. Based on the deleted target point cloud in a unified coordinate system, the spatial distribution of the acquired target point cloud is traversed across the entire domain. The changes in point cloud distribution density of adjacent spatial positions within the target point cloud are statistically analyzed segment by segment. The spatial direction in which the density changes continuously from large to small is identified, and each complete and uninterrupted continuous spatial direction is determined as a distribution direction of the target point cloud. The principle is that the vehicle body is a continuous entity, and its point cloud distribution is continuous. Moreover, the point cloud density of different parts of the vehicle body is different. For example, the point cloud density is large at the position of the vehicle body closer to the vehicle 4 and small at the position farther away from the vehicle 4. The direction of extension of the vehicle body can be identified by the density change. The specific implementation method is as follows: The area covered by the target point cloud on the horizontal plane is divided into equally spaced grids to obtain multiple continuous grids. The principle is that grid division discretizes the continuous point cloud region into statistically significant units, facilitating accurate point cloud density calculation. The grid interval can be set according to the point cloud density, typically 5cm × 5cm. This makes density statistics more targeted and avoids insignificant density changes caused by global statistics. The number of point clouds contained in each grid is counted sequentially, and the difference in the number of point clouds in adjacent spatial grids is compared to obtain the point cloud distribution density changes in adjacent spatial locations within the target point cloud set. The principle is that the number of point clouds is positively correlated with density, and the difference in the number of point clouds in adjacent grids can intuitively reflect the density change trend. The expression for differential calculation is as follows: (3) In the formula, For the first grid Point cloud density within; This represents the density difference between adjacent grids.

[0033] This allows for the accurate capture of continuous changes in point cloud density, providing a reliable basis for distribution direction identification. For example, if there are 12 point clouds in a certain grid, the number of point clouds in adjacent grids... For 10, There are 11 grids, and the density difference between adjacent grids is +1, so they show a continuous decreasing trend, which can be used to identify this direction as one of the distribution directions of the point cloud.

[0034] When two sets of independent distribution directions are identified, two corresponding target functions are obtained by fitting a linear function to each direction. The angle between the distributions of the two target functions is calculated. When the angle is within a preset range, the cluster of target points is identified as the first cluster. The principle is that the two main extension directions of the vehicle ahead in adjacent lane 2 (i.e., the driving direction and the lateral direction away from vehicle 4) are nearly perpendicular. The distribution direction can be quantified by fitting a linear function, and the angle calculation determines whether it is the side and rear of the same preceding vehicle. The angle calculation is as follows: Let the direction vectors obtained by fitting a straight line to the two main distribution directions identified from the target point cloud be... and Then the included angle satisfy: (4) Wherein, direction vector and It is calculated using the following formula: (5) In the formula, The direction vector of a target point aggregate along a certain distribution direction; This represents the number of target point clouds participating in the fitting along this distribution direction; For the direction of this distribution, the first The two-dimensional coordinates of the target point cloud on the horizontal plane; For all target point clouds in this distribution direction The average value of the coordinates; For all target point clouds in this distribution direction The average value of the coordinates.

[0035] in: (6) (7) S104, perform piecewise straight line fitting on the first point cloud, and perform straight line fitting on the second and third point clouds; the straight line fitting includes: in the corresponding target point cloud, select a preset number of target position points located on the outermost side of the vehicle in front, and fit the target position points into a reference straight line of the vehicle in front; map the points located at the end positions in the corresponding target point cloud to the reference straight line as endpoints; and generate the warning line of the vehicle in front based on the reference straight line with the endpoints as constraints.

[0036] Specifically, the first point set is fitted piecewise, while the second and third point sets are fitted directly as straight lines. The principle is based on the differences in the relative positions and attitudes of the preceding vehicle and the current vehicle 4 corresponding to different point sets. A differentiated fitting method is adopted to ensure that the reference straight line can accurately match the outline features of the preceding vehicle, providing a reliable benchmark for the subsequent generation of warning lines. At the same time, it takes into account both fitting accuracy and computational efficiency, avoiding outline deviations caused by a single fitting method.

[0037] The linear fitting process includes: selecting a predetermined number of target position points located on the outermost edge of the preceding vehicle from the corresponding target point cloud, and fitting these target position points to form a reference straight line for the preceding vehicle. The core principle of selecting the outermost target position points is that the outermost point cloud of the preceding vehicle directly corresponds to the vehicle's contour boundary. Selecting a predetermined number of points (e.g., 10-15) avoids fitting errors caused by deviations from a single outermost point, while reducing the amount of fitting computation, balancing fitting accuracy and computational efficiency, and ensuring that the reference straight line accurately outlines the preceding vehicle's contour. The fitting uses the least squares method, minimizing the sum of squared distances from the target position points to the fitted straight line to obtain a reference straight line that fits the preceding vehicle's contour. This method is computationally simple, highly resistant to interference, and can effectively avoid fitting deviations caused by slight discrepancies in the point cloud, ensuring the stability of the reference straight line. The calculation process is as follows: Let the selected set of outer target locations be... The equation of the line to be fitted is ,in, Let be the slope of the line. The least squares method finds the optimal solution by minimizing the sum of squared residuals from all sample points to the fitted line. and The formula for calculating the sum of squared residuals is: (8) In the formula, For the sum of squared residuals, For the first The coordinates of the target location points on the horizontal plane. The optimization goal of formula (8) is to find the optimal slope. and intercept This makes the sum of squares of the residuals... Obtain the minimum value.

[0038] To achieve the objective of formula (8), respectively, for and Find the partial derivatives, set them to 0, and we get... , and Relationship: (9) (10) The process maps points located at the ends of the target point cloud to a reference line as endpoints. The principle is that the end points in the target point cloud correspond to the front, rear, or end of the vehicle ahead. Mapping them to the reference line accurately locates the start and end range of the vehicle's outline, clearly defining the effective range of the reference line and preventing it from exceeding the actual outline of the vehicle ahead. This provides precise boundary constraints for generating the warning line, ensuring that the warning line completely covers the danger zone of the vehicle ahead. The mapping process relies on the spatial uniformity of a unified coordinate system. The vertical distance from the end point to the reference line is calculated, and this point is projected vertically onto the reference line; the projected point is the endpoint. This method allows for rapid mapping without changing the relative spatial position of the endpoints, further ensuring the accuracy of endpoint positioning. Then, using the endpoints as constraints, the warning line for the vehicle ahead is generated based on the reference line. In other words, using these endpoints as the endpoints of the warning line makes the warning line a line segment that can measure the length or width of the vehicle ahead.

[0039] The step of selecting a predetermined number of target location points located on the outermost edge of the preceding vehicle specifically includes: In the second point cloud set, calculating the distance from each point on the lateral side of vehicle 4 to the origin of the coordinate system, and selecting the predetermined number of points with the smallest distance as the outermost target location points. Since the second point cloud set corresponds to the preceding vehicle in the adjacent lane 2 that is parallel or closely following, its point cloud is distributed along the driving direction of vehicle 4. The lateral side of vehicle 4 is the y-axis direction of the coordinate system. The point cloud on the horizontal plane only contains x (driving direction) and y coordinates. The lateral distance to the origin is the y-coordinate value, so no additional distance calculation is required, further reducing the computational load. The principle is that the outermost point of this type of preceding vehicle is closest to the origin on the lateral side of vehicle 4. Selecting this type of point can accurately locate the lateral boundary of the preceding vehicle, simplifying the calculation process, improving response speed, and ensuring that the target location point fits the outer contour of the preceding vehicle.

[0040] In the third point cloud set, calculate the distance from each point in the driving direction of vehicle 4 to the origin of the coordinate system, and select the preset number of points with the smallest distance as the outermost target position points; since the third point cloud set corresponds to the vehicle directly in front in the same lane, its point cloud is distributed laterally along vehicle 4, and the driving direction of vehicle 4 is the x-axis direction of the coordinate system, the x-coordinate of the point cloud on the horizontal plane is the distance from the origin in the driving direction, which does not require additional calculation and can reduce the amount of calculation.

[0041] S105, real-time detection of the position and deflection angle of each warning line; if the preceding vehicle corresponding to the first point cluster meets the preset lane change rules, the vehicle controller of this vehicle 4 is driven to control the vehicle 4 to decelerate or brake according to the distance; if the preceding vehicle corresponding to the second point cluster meets the preset lane change rules, the vehicle controller is driven to brake suddenly; if the distance between the warning line of the preceding vehicle corresponding to the third point cluster and this vehicle 4 is less than the first preset distance, the vehicle controller is driven to decelerate or brake.

[0042] Specifically, the real-time detection of the position and deflection angle of each warning line relies on the spatial uniformity of a unified coordinate system. This is achieved by combining the real-time updated target point clusters from the vehicle radar 3 with image data collected by the vehicle camera. The spatial coordinates of each warning line are obtained in real-time through coordinate comparison, and the angle between the warning line and the direction of travel of the vehicle 4 (i.e., the deflection angle) is calculated, tracking the dynamic changes of the warning lines in real time. If the vehicle ahead of the first point cluster meets the preset lane-changing rules, the vehicle controller of the vehicle 4 will control the vehicle 4 to decelerate or brake based on the distance. If the first point cluster corresponds to the vehicle ahead in the adjacent lane 2, the threat posed by such a vehicle changing lanes is relatively mild. Therefore, the vehicle 4 is controlled in stages based on the actual distance between the preceding vehicle and the vehicle 4. The principle is that the vehicle radar 3 calculates the straight-line distance between the preceding vehicle's warning line and the vehicle 4 in real time, and combines this with the vehicle 4's current speed. The vehicle controller then matches the corresponding deceleration magnitude or braking force. For example, a slight deceleration occurs when the distance is greater than 50 meters, a moderate deceleration occurs when the distance is less than 30 meters, and braking is triggered when the distance is less than 15 meters.

[0043] If the preceding vehicle in the second lane meets the preset lane change rules, the vehicle controller will apply emergency braking. The preceding vehicle in the second lane is parallel or closely following in the adjacent lane 2. When such a preceding vehicle changes lanes, the lateral distance to the vehicle 4 is close, and the risk of collision is high. Emergency braking can shorten the braking distance to the maximum extent. The principle is that after the vehicle controller receives the trigger signal, it immediately cuts off the acceleration command and starts the maximum braking force. At the same time, it links the ABS anti-lock braking system to avoid wheel lock-up and sideslip, thereby quickly responding to sudden lane change risks and reducing the probability of collision.

[0044] If the distance between the warning line corresponding to the third convergence point and the vehicle 4 is less than the first preset distance, the vehicle controller will decelerate or brake. The first preset distance for the vehicle directly ahead in the same lane corresponding to the third convergence point can be dynamically adjusted according to the vehicle 4's speed. For example, the preset distance is 30 meters at 60 km / h and 50 meters at 100 km / h. The principle is that when the distance between the two vehicles is less than this threshold, there is a risk of a rear-end collision. Decelerating or braking increases the safe distance, thus avoiding the risk of a rear-end collision in the same lane and adapting to the safety distance requirements at different speeds.

[0045] The preset lane-changing rules include: The angle between the warning line of the preceding vehicle and the direction of travel of the current vehicle (4) is greater than a preset angle, which can be set to 15°. The principle is that when the preceding vehicle changes lanes, its body will deflect, causing the warning line to deflect at a larger angle. When the angle exceeds 15°, it can be determined that the preceding vehicle intends to change lanes; the lane-changing trend can be quickly identified through angle changes. Alternatively, if the image data detects that the turn signal of the preceding vehicle is continuously illuminated towards the current vehicle's lane 1 within a preset time period, the preset time is set to 2 seconds. The principle is that the preceding vehicle usually turns on its turn signal in advance before changing lanes; continuous illumination for 2 seconds can rule out accidental activation. Combined with the image recognition function of the onboard camera, the turn signal status can be accurately detected. Alternatively, the warning line will shift towards the current vehicle's lane 1 by a second preset distance within a preset time period, with a preset time of 1 second and a second preset distance of 40cm. The principle is that when the preceding vehicle changes lanes, it will move towards the current vehicle's lane 1, causing the warning line to shift synchronously. If the shift reaches 40cm and lasts for 1 second, the lane-changing action can be confirmed. The lane-changing behavior is quantified through positional shift, avoiding misjudgments and ensuring the reliability of the decision.

[0046] In embodiments of the present invention, for complex road conditions in intelligent driving of electric vehicles, multi-source perception fusion and point cloud data optimization processing are used to achieve accurate recognition of the intention of the vehicle in front, and this serves as the core support for electric vehicle control, significantly improving the safety and real-time decision-making of intelligent driving in complex road conditions. Furthermore, relying on the vehicle-mounted radar 3 and camera with a unified coordinate system to acquire point cloud and image data of the vehicle in front, the three-dimensional point cloud is simplified into two-dimensional data by constructing a horizontal plane, significantly reducing the computational load and meeting real-time decision-making requirements. Then, based on the point cloud distribution characteristics, the scene in which the vehicle in front is located is distinguished, and a warning line for the vehicle in front is fitted and generated to accurately lock the outline and dynamic changes of the vehicle in front. Finally, based on the position, deflection angle, and turn signal of the warning line, the lane-changing intention of the vehicle in front is accurately determined, and a differentiated electric vehicle control strategy is executed accordingly: the vehicle in front of the adjacent lane 2 is decelerated or braked in stages according to distance; emergency braking is triggered for the vehicle in front of the parallel adjacent lane 2; and deceleration and braking are controlled according to a safe distance for the vehicle directly in front in the same lane.

[0047] This embodiment improves response speed through lightweight data processing and effectively avoids the risks of rear-end collisions and scrapes through precise and hierarchical electric vehicle control, providing a reliable guarantee for the safety decision-making of intelligent driving of electric vehicles in complex road conditions.

[0048] Example 2 Based on the method in Example 1, when encountering a dangerous vehicle, such as a truck carrying wooden stakes exceeding its own length under certain road conditions, the existing model struggles to identify the vehicle as dangerous and tends to treat it as a single vehicle. Therefore, to address this issue, the following method is adopted: First, when the target point cloud is distributed along three or more directions, it is identified as the fourth point cloud; the fourth point cloud is fitted with a multi-segment continuous straight line; the multi-segment continuous straight line fitting includes: fitting all points in the fourth point cloud to form a piecewise linear function; Secondly, in the piecewise linear function, sub-functions 7 whose slopes are within the preset range of the lateral slope error of the vehicle 4 are selected, and the distance between the two farthest sub-functions 7 in the driving direction is calculated. If the distance is greater than the threshold, the vehicle is judged as a dangerous vehicle. Third, take the line segment where the nearest sub-function 7 is located in the direction of travel as the warning line, control the vehicle 4 to maintain a third preset distance from the warning line in the direction of travel, or control the vehicle 4 to accelerate and overtake the vehicle in front corresponding to the fourth point gathering.

[0049] Specifically, when the target point cloud is distributed along three or more directions, it is considered a fourth point cloud. The principle is that the outline of a conventional vehicle typically only shows two or fewer main extension directions in its horizontal projection, corresponding to the side and rear of the vehicle. When the point cloud distribution reaches three or more directions, it indicates that the vehicle has additional protruding structures besides its basic body outline, such as the extra-long wooden stakes loaded behind a truck. This structure forms an independent distribution direction in the horizontal point cloud that is inconsistent with the vehicle's extension direction. Figure 8 and Figure 9 As shown, this triggers the fourth point of the cloud-based classification mechanism, providing a basic classification basis for subsequent hazard identification of oversized loads.

[0050] The principle behind fitting the fourth point cloud with multiple continuous straight lines is that the point cloud of such non-standard contour vehicles cannot be accurately described by a single reference straight line. It requires segmented approximation based on the spatial distribution characteristics of the point cloud itself to preserve contour transitions and highlight details, avoiding the loss of contour information due to overall fitting. Multi-segment continuous straight line fitting includes fitting all points within the fourth point cloud to form a piecewise linear function, such as... Figure 9 As shown, the specific implementation involves performing a piecewise linear approximation of the area covered by the fourth point cloud on the horizontal plane. The segment nodes can be determined based on the location of the discontinuities in the point cloud density. The point cloud subsets between each node are fitted with least squares straight lines, and the straight line segments are connected at the nodes to form a complete piecewise linear function. This function can reflect the contour changes of the vehicle and its cargo in the horizontal plane projection. For example, the edge of the truck bed at the rear of the truck 6, the end face of the wooden stake, and the side of the wooden stake will be represented by sub-functions 7 with different slopes in the piecewise linear function. The overall contour presents a stepped or broken line feature.

[0051] Since the fitted line is a straight line, the point cloud distribution consists of continuous straight line segments. There will be significant discontinuities in the point cloud density corresponding to different straight line segments (e.g., at the connection between the truck body and the wooden stake, the point cloud density will suddenly decrease or increase). Therefore, the relative difference in point cloud density between adjacent grids is used as the criterion. When the relative difference exceeds a preset threshold, it indicates a discontinuity in the point cloud distribution at that location, corresponding to different straight line fitting segments, and is determined as a segmented node. Its expression is as follows: (11) In the formula, For the first Point cloud density of each grid; For the first Point cloud density of +1 grid; This is a preset threshold.

[0052] In the piecewise linear function, sub-functions 7 whose slopes differ from the lateral slope of vehicle 4 within a preset range are selected. The preset range can be set to an absolute value of the slope difference not exceeding 0.3. The principle is that the rear end face of the vehicle and the cut end face of the wooden stake are usually nearly parallel to the lateral direction of vehicle 4. Slope matching can quickly filter out line segments representing the rear end and the end face of the wooden stake, excluding line segments representing the sides of the vehicle body and other inclined contours, thus achieving preliminary positioning of the end face features. Based on this, the distance between the two furthest sub-functions 7 in the driving direction is calculated. The principle is that the sub-function 7 corresponding to the rear end face and the sub-function 7 corresponding to the furthest end face of the wooden stake are furthest apart in the longitudinal direction of the vehicle. This distance directly reflects the length of the load exceeding the vehicle body. If the distance is greater than a threshold, for example, a threshold set at 100 cm more than the standard cargo box length of a regular truck 6, it is judged as a dangerous vehicle. This distinguishes between normal loading and over-length loading, avoiding misjudging a regular truck 6 as a dangerous vehicle, and accurately identifying abnormal vehicles that may pose a collision risk due to protruding wooden stakes.

[0053] The line segment containing the sub-function 8 closest to vehicle 4 in the direction of travel is used as the warning line. Vehicle 4 is controlled to maintain a third preset distance from the warning line in the direction of travel, or vehicle 4 is controlled to accelerate and overtake the vehicle in front corresponding to the fourth point cluster. The principle is that the sub-function 8 closest to vehicle 4 usually corresponds to the rear end face of the vehicle in front. Using it as the warning line can directly reflect the longitudinal position of the rear of the vehicle in front of vehicle 4. By maintaining the third preset distance, for example, set to 40 meters according to the speed of vehicle 4, rear-end collisions caused by sudden braking of the vehicle in front or protruding cargo can be effectively prevented. When the road conditions permit and it is necessary to avoid the risk of continuous following, vehicle 4 is controlled to accelerate and overtake the vehicle in front to get out of the dangerous following state and improve the driving safety margin.

[0054] The third preset distance From the reaction distance ( Braking distance and safety redundancy distance Composition. It can dynamically adjust according to the vehicle's real-time speed to ensure a sufficient safe distance at different speeds, adapting to the following control requirements of dangerous vehicles. The specific calculation formula is as follows: (12) In the formula, This is the current speed of vehicle 4; This is the vehicle's 4-stage braking reaction time; This is the maximum braking acceleration of the vehicle.

[0055] Example 3 Based on the method in Example 1, after obtaining the warning line, simply adding the vertical height to the warning line is sufficient to obtain the basic model of the vehicle in front, thereby improving the speed at which the vehicle's infotainment system displays real-time vehicles. The specific method is as follows: The height of the preceding vehicle is calculated based on the point cloud data of each preceding vehicle; For the vehicle in front corresponding to the first point cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side and the wide side of the vehicle in front. Based on the height of the vehicle in front and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front of the second cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side of the vehicle in front. Based on the height of the vehicle in front, the preset assumed width and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front corresponding to the third point cluster, the length of the warning line of the vehicle in front is used as the constraint of the width of the vehicle in front. Based on the height of the vehicle in front, the preset assumed length and the preset vehicle model, a real-time model of the vehicle in front is established. All real-time models are displayed on the vehicle's infotainment screen based on their corresponding positions.

[0056] Example 4 like Figure 10 As shown, based on the same inventive concept, this embodiment provides a complex road condition intention recognition and decision-making system for intelligent driving of electric vehicles, including: The perception and acquisition module 201 is used to acquire the driving direction and current position of the vehicle 4; acquire valid point cloud data of the vehicle in front through the vehicle radar 3; and acquire image data of the vehicle in front through the vehicle camera. The point cloud filtering module 202 is used to filter the point 5 closest to the current vehicle 4 from the point cloud data of each preceding vehicle as the reference point of the corresponding preceding vehicle, and to establish the horizontal plane of the preceding vehicle in the coordinate system with the reference point; and to filter the point cloud data on the preceding vehicle located on the horizontal plane as the target point cloud set. The point cloud classification module 203 is used to identify a target point cloud as a first point cloud when the target point cloud is distributed along two directions and the two directions are within a preset angle range; to identify a target point cloud as a second point cloud when the target point cloud is distributed along the driving direction of the vehicle 4; and to identify a target point cloud as a third point cloud when the target point cloud is distributed laterally along the vehicle 4. The fitting module 204 is used to perform piecewise straight line fitting on the first point cloud and straight line fitting on the second and third point clouds. The straight line fitting includes: in the corresponding target point cloud, selecting a preset number of target position points located on the outermost side of the vehicle in front, and fitting the target position points into a reference straight line of the vehicle in front; mapping the points located at the end positions in the corresponding target point cloud to the reference straight line as endpoints; and generating a warning line for the vehicle in front based on the reference straight line with the endpoints as constraints. The decision module 205 is used to detect the position and deflection angle of each warning line in real time. If the vehicle in front of the first point cluster meets the preset lane change rules, the vehicle controller of the vehicle 4 is driven to decelerate or brake according to the distance. If the vehicle in front of the second point cluster meets the preset lane change rules, the vehicle controller is driven to brake suddenly. If the distance between the warning line of the vehicle in front of the third point cluster and the vehicle 4 is less than the first preset distance, the vehicle controller is driven to decelerate or brake. The preset lane-changing rules include: the angle between the warning line of the preceding vehicle and the direction of travel of the vehicle 4 is greater than a preset angle, or the turn signal in the image data is continuously illuminated in the direction of the vehicle 4 lane 1 within a preset time, or the warning line is shifted to the direction of the vehicle 4 lane 1 by a second preset distance within a preset time.

[0057] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0058] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles, characterized in that, The method includes onboard radar and onboard cameras mounted on electric vehicles and operating in a unified coordinate system. S101: Obtain the vehicle's driving direction and current position; acquire valid point cloud data of the vehicle in front via vehicle radar; acquire image data of the vehicle in front via vehicle camera; S102, from the point cloud data of each of the preceding vehicles, select the point closest to the current vehicle as the reference point of the corresponding preceding vehicle, and establish the horizontal plane of the preceding vehicle in the coordinate system with the reference point; select the point cloud data of the preceding vehicle located on the horizontal plane as the target point cloud set; S103, when the target points are clustered along two directions and the two directions are within a preset angle range, they are identified as a first cluster; when the target points are clustered along the vehicle's driving direction, they are identified as a second cluster; when the target points are clustered along the vehicle's lateral direction, they are identified as a third cluster. S104, perform piecewise straight line fitting on the first point cloud, and perform straight line fitting on the second point cloud and the third point cloud; the straight line fitting includes: in the corresponding target point cloud, selecting a preset number of target position points located on the outermost side of the vehicle in front, and fitting the target position points into a reference straight line for the vehicle in front; mapping the points located at the end positions in the corresponding target point cloud to the reference straight line as endpoints; using the endpoints as constraints, generating a warning line for the vehicle in front based on the reference straight line; S105, Real-time detection of the position and deflection angle of each warning line; If the preceding vehicle corresponding to the first point cluster meets the preset lane change rules, the vehicle controller of this vehicle is driven to control the vehicle to decelerate or brake according to the distance; If the preceding vehicle corresponding to the second point cluster meets the preset lane change rules, the vehicle controller is driven to brake suddenly; If the distance between the warning line of the preceding vehicle corresponding to the third point cluster and this vehicle is less than the first preset distance, the vehicle controller is driven to decelerate or brake. The preset lane-changing rules include: the angle between the warning line of the preceding vehicle and the direction of travel of the current vehicle is greater than a preset angle, or the turn signal of the preceding vehicle towards the direction of the current vehicle's lane is continuously illuminated in the image data within a preset time, or the warning line is shifted towards the direction of the current vehicle's lane by a second preset distance within a preset time.

2. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, The step of selecting the point closest to the current vehicle as the reference point for the corresponding preceding vehicle specifically includes: Based on the time-of-flight ranging principle of the vehicle-mounted radar, multiple echo points with the shortest time of reflection are obtained from all the point cloud data of the preceding vehicle. Multiple echo points are clustered according to their spatial distribution density, and the point located at the center of the clustered area is extracted as the reference point.

3. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, The step of establishing the horizontal plane of the preceding vehicle in the coordinate system based on the reference point specifically includes: Centered on the reference point, a virtual plane with a normal vector parallel to the direction of the vehicle's gravity is generated in the unified coordinate system; the virtual plane is defined as the horizontal plane of the vehicle in front.

4. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, The step of identifying a target point cluster as the first point cluster when the target point cluster is distributed along two directions and the two directions are within a preset angle range specifically includes: Under the unified coordinate system, the spatial distribution of the acquired target point cloud is traversed across the entire domain. The changes in point cloud distribution density of adjacent spatial positions within the target point cloud are statistically analyzed segment by segment. Spatial directions in which the density changes continuously from large to small are identified. Each complete and uninterrupted continuous spatial direction is determined as a distribution direction of the target point cloud. When two sets of independent distribution directions are identified, two corresponding target functions are obtained by fitting the two directions with a linear function; the distribution angle between the two target functions is calculated, and when the distribution angle is within a preset range, the target point set is identified as the first point set.

5. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 4, characterized in that, Before performing a global spatial distribution traversal on the obtained target point cloud, the process also includes: The vertical coordinate information of the target point cloud is deleted, and only the two-dimensional coordinates on the horizontal plane are used for subsequent direction identification and angle calculation.

6. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 4, characterized in that, The step of statistically analyzing the changes in point cloud distribution density at adjacent spatial locations within the target point cloud set segment by segment specifically includes: The area covered by the cluster of target points on the horizontal plane is divided into multiple continuous grids at equal intervals. The number of point clouds contained in each grid is counted sequentially, and the difference in the number of point clouds in adjacent spatial grids is compared to obtain the change in point cloud distribution density at adjacent spatial locations within the target point cloud set.

7. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, The step of selecting a predetermined number of target location points located on the outermost edge of the preceding vehicle specifically includes: In the second point cloud, calculate the distance from each point in the vehicle's driving direction to the origin of the coordinate system, and select the preset number of points with the smallest distance as the outermost target position points; In the third point cloud, the distance from each point on the lateral side of the vehicle to the origin of the coordinate system is calculated, and a preset number of points with the smallest distance are selected as the outermost target position points.

8. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, Also includes: When the target point cluster is distributed along three or more directions, it is identified as a fourth point cluster; The fourth point cluster is fitted with multiple continuous straight lines; The multi-segment continuous straight line fitting includes: fitting all points within the fourth point set to form a piecewise linear function; In the piecewise linear function, sub-functions whose slope and the lateral slope error of the vehicle are within a preset range are selected. The distance between the two farthest sub-functions in the driving direction is calculated. If the distance is greater than a threshold, the vehicle is judged to be dangerous. The line segment containing the sub-function closest to the vehicle in the direction of travel is used as the warning line. The vehicle is controlled to maintain a third preset distance from the warning line in the direction of travel, or the vehicle is controlled to accelerate and exceed the preceding vehicle corresponding to the fourth point cluster.

9. The method for complex road condition intent recognition and decision-making in intelligent driving of electric vehicles according to claim 1, characterized in that, Also includes: The height of the preceding vehicle is calculated based on the point cloud data of each preceding vehicle; For the vehicle in front corresponding to the first point cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side and the wide side of the vehicle in front. Based on the height of the vehicle in front and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front corresponding to the second point cluster, the length of the warning line of the vehicle in front is used as the constraint of the long side of the vehicle in front. Based on the height of the vehicle in front, the preset assumed width and the preset vehicle model, a real-time model of the vehicle in front is established. For the vehicle in front corresponding to the third point cluster, the length of the warning line of the vehicle in front is used as the constraint of the width of the vehicle in front. Based on the height of the vehicle in front, the preset assumed length and the preset vehicle model, a real-time model of the vehicle in front is established. The real-time model is displayed on the vehicle's infotainment screen.

10. A complex road condition intent recognition and decision-making system for intelligent driving of electric vehicles, based on the complex road condition intent recognition and decision-making method for intelligent driving of electric vehicles as described in claim 1, characterized in that, include: The perception and acquisition module is used to obtain the vehicle's driving direction and current position; and to acquire effective point cloud data of the vehicle in front through the vehicle-mounted radar. Acquire image data of the vehicle in front using an in-vehicle camera; The point cloud filtering module is used to filter the point closest to the current vehicle from the point cloud data of each preceding vehicle as the reference point of the corresponding preceding vehicle, and to establish the horizontal plane of the preceding vehicle in the coordinate system with the reference point; and to filter the point cloud data of the preceding vehicle located on the horizontal plane as the target point cloud set. The point cloud classification module is used to identify a target point cloud as a first point cloud when the target point cloud is distributed along two directions and the two directions are within a preset angle range; to identify a target point cloud as a second point cloud when the target point cloud is distributed along the vehicle's driving direction; and to identify a target point cloud as a third point cloud when the target point cloud is distributed laterally along the vehicle. The fitting module is used to perform piecewise straight line fitting on the first point cloud, and to perform straight line fitting on the second point cloud and the third point cloud; the straight line fitting includes: in the corresponding target point cloud, selecting a preset number of target position points located on the outermost side of the preceding vehicle, and fitting the target position points into a reference straight line of the preceding vehicle; mapping the points located at the end positions in the corresponding target point cloud to the reference straight line as endpoints; and generating a warning line for the preceding vehicle based on the reference straight line with the endpoints as constraints. The decision module is used to detect the position and deflection angle of each warning line in real time; if the vehicle in front of the first point cluster meets the preset lane change rules, the vehicle controller of the vehicle is driven to control the vehicle to decelerate or brake according to the distance; if the vehicle in front of the second point cluster meets the preset lane change rules, the vehicle controller is driven to brake suddenly; if the distance between the warning line of the vehicle in front of the third point cluster and the vehicle is less than the first preset distance, the vehicle controller is driven to decelerate or brake. The preset lane-changing rules include: the angle between the warning line of the preceding vehicle and the direction of travel of the current vehicle is greater than a preset angle, or the turn signal in the image data is continuously illuminated in the direction of the current vehicle's lane within a preset time, or the warning line is shifted to the direction of the current vehicle's lane by a second preset distance within a preset time.

Images