Intelligent driving track determination method and device based on visual model and medium

Abstract

The invention relates to the technical field of driving track generation, and discloses an intelligent driving track determination method and device based on a visual model, and a medium, and the method comprises the steps: obtaining a road visual image of a preset direction of a target vehicle, carrying out the semantic segmentation of the road visual image, and obtaining a plurality of pieces of semantic visual information, extracting weak geometric features of the road visual image, constructing a scene semantic graph according to the semantic visual information and the weak geometric features, carrying out central axis dynamic identification on a passable area of the scene semantic graph, generating a plurality of candidate driving tracks according to the identified central axis, analyzing the passability of the candidate driving tracks, and determining the passability of the candidate driving tracks. And performing optimal track screening on the candidate driving tracks according to the passable probability to obtain a target driving track of the target vehicle. According to the invention, when a lane line is unavailable or unreliable, effective track determination can be carried out through other visual elements.

Description

Technical Field

[0001] The present invention relates to the technical field of driving trajectory generation, and in particular to a method, device and medium for determining an intelligent driving trajectory based on a visual model. Background Art

[0002] With the continuous development of intelligent driving technology, trajectory planning methods based on visual perception have become an important research direction in autonomous driving systems due to their low cost, rich information content, and high integration flexibility. Existing visual trajectory planning methods mainly rely on lane detection, semantic segmentation, and end-to-end neural network models to achieve vehicle path prediction and control. These methods generally assume that the road structure is clear and the lane lines are intact. The system uses lane boundaries and centerlines extracted from the image as the basic reference for trajectory generation. However, in actual road environments, especially in "weak lane" or "no lane" scenarios such as urban-rural fringe areas, temporary construction areas, snowy days, or at night, lane lines may appear blurred, obscured, missing, or even misleading, seriously affecting the stability of the visual perception module and the safety of trajectory decision-making.

[0003] Furthermore, existing methods mostly rely on regularized trajectory templates or predefined center lane lines. These methods lack sufficient adaptability and robustness for complex topologies or unstructured roads (such as roundabouts, plazas, and parking lots), and are unable to make effective path decisions in the absence of clear lane guidance. Although some studies have attempted to incorporate map data for assisted planning, these approaches perform poorly in dynamically changing road environments due to limitations in map accuracy and timeliness. Therefore, there is an urgent need for a new method that can effectively predict and control trajectories using other visual elements when lane lines are unavailable or unreliable, thereby improving the system's generalization and driving safety in weakly structured road environments.

[0004] In summary, in the current existing technologies, when lane lines are unavailable or unreliable, there is a problem that it is difficult to rely on other visual elements to effectively determine the trajectory. Summary of the Invention

[0005] The present invention provides a method, device and medium for intelligent driving trajectory determination based on a visual model, the main purpose of which is to solve the problem that when lane lines are unavailable or unreliable, it is difficult to rely on other visual elements for effective trajectory determination.

[0006] In a first aspect, to achieve the above-mentioned objectives, the present invention provides a method for determining an intelligent driving trajectory based on a visual model, comprising: Acquire a road visual image in a preset direction of the target vehicle, perform semantic segmentation on the road visual image, and obtain a plurality of semantic visual information; Extracting weak geometric features of the road visual image, and constructing a scene semantic graph based on the semantic visual information and the weak geometric features; Dynamically identifying a central axis of a traversable area of the scene semantic graph, and generating a plurality of candidate driving trajectories based on the identified central axis; Analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability; The candidate driving trajectories are screened for optimal trajectories according to the passable probabilities to obtain a target driving trajectory of the target vehicle.

[0007] In a second aspect, the present invention further provides an electronic device, comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned method for determining an intelligent driving trajectory based on a visual model.

[0008] In a third aspect, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned method for determining an intelligent driving trajectory based on a visual model.

[0009] The present invention obtains a road visual image of a preset direction of a target vehicle, performs semantic segmentation on the road visual image, and obtains a number of semantic visual information. By combining shallow semantic features, high-resolution details, and contextual features enhanced by hole convolution, and adopting a jump connection fusion mechanism, the collaborative expression of local details and global semantics is effectively achieved, thereby significantly improving the segmentation accuracy and robustness, and providing more accurate and stable semantic visual information support for tasks such as target recognition, path extraction, and environmental perception in road scenes. The weak geometric features of the road visual image are extracted, and a scene semantic graph is constructed based on the semantic visual information and the weak geometric features. Fine gradient calculation and convolution coding ensure the accurate capture of geometric structure features, and the key channels are strengthened in conjunction with the attention mechanism, thereby improving the pertinence and effectiveness of feature expression. The geometrically guided semantic annotation graph adjustment and deconvolution recovery are based on the geometry-guided semantic annotation graph, which enhances the spatial consistency and detail integrity of the semantic structure. The geometrically guided path and semantic nodes are fused into a scene semantic graph, which is beneficial to improving the accuracy of target recognition, path planning, and scene understanding in road environments. The method dynamically identifies the central axis of the traversable area in the scene semantic graph and generates several candidate driving trajectories based on the identified central axis. By dynamically identifying the central axis of the traversable area in the scene semantic graph and generating smooth and reasonable candidate driving trajectories based on the axis, the method can adapt to complex road conditions and changes in traffic direction in different scenarios and accurately model the main traffic trend. The method analyzes the traversability of the candidate driving trajectories to obtain a traversability probability. The candidate driving trajectories are optimally screened based on the traversability probability to obtain the target driving trajectory of the target vehicle. The traversability probability is used to eliminate high-risk or unreasonable trajectories to ensure that the candidate path set has basic traffic capability and stability. The method generates a clear driving intention in combination with the vehicle's navigation goal and finely matches the retained trajectories based on the intention to ensure that the final selected target driving trajectory is not only safe and reliable but also accurately responds to driving needs and road environment changes. The method effectively solves the problem of difficulty in relying on other visual elements for effective trajectory determination when lane lines are unavailable or unreliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0011] Figure 1 A flowchart of a method for determining an intelligent driving trajectory based on a visual model provided by one embodiment of the present invention; Figure 2A schematic structural diagram of an electronic device for implementing a method for determining an intelligent driving trajectory based on a visual model according to an embodiment of the present invention; Figure 3 This is another structural diagram of an electronic device for implementing a method for determining an intelligent driving trajectory based on a visual model provided by an embodiment of the present invention.

[0012] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0013] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, and to fully understand and implement how the present disclosure applies technical means to solve technical problems and achieve the corresponding technical effects, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. The embodiments of the present disclosure and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0014] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0015] An embodiment of the present application provides a method for determining an intelligent driving trajectory based on a visual model, and the execution subject of the method for determining an intelligent driving trajectory based on a visual model includes but is not limited to a server, a terminal, etc. that can be configured to execute at least one of the electronic devices of the system provided by the embodiment of the present application. In other words, the method for determining an intelligent driving trajectory based on a visual model can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0016] Reference Figure 1 FIG2 is a flow chart of a method for determining an intelligent driving trajectory based on a visual model according to an embodiment of the present invention. In this embodiment, the method for determining an intelligent driving trajectory based on a visual model includes: S1. Obtain a road visual image of a preset direction of a target vehicle, perform semantic segmentation on the road visual image, and obtain a plurality of semantic visual information.

[0017] In this embodiment of the present invention, a forward-looking camera captures real-time road images in a preset direction ahead of the target vehicle while the vehicle is in motion. After distortion correction and image enhancement, a deep neural network with multi-scale feature fusion is used to perform semantic segmentation on the processed road images to obtain semantic visual information. This semantic visual information includes, but is not limited to, road surface areas, curbs, dynamic objects (such as pedestrians and vehicles), traffic signs, and obstacle boundaries.

[0018] Specifically, the road visual image is semantically segmented to obtain a plurality of semantic visual information, including: Performing distortion correction on the road visual image to obtain a corrected visual image; performing edge enhancement processing on the corrected visual image to obtain an enhanced visual image; extracting shallow semantic features and high-resolution features of the enhanced visual image; Obtaining a dilation rate and a dilation convolution kernel corresponding to the dilation rate, and performing channel splicing on the enhanced visual image using the dilation convolution kernel to obtain a visual convolution feature; The shallow semantic features, the high-resolution features and the visual convolutional features are skip-connected and fused to obtain semantic visual information.

[0019] In detail, based on the camera's intrinsic parameter matrix and distortion coefficients, the distortion model is used to perform reverse mapping on the original visual image to eliminate the geometric deformation caused by lens distortion. The mapped image pixels are reconstructed using image resampling methods such as bilinear interpolation, and finally a corrected visual image with more realistic geometric structure and more accurate edge contour is generated.

[0020] Edge detection operators such as Sobel, Laplacian or Canny are used to extract significant edge information in the image. The contrast and detail features of the edge area in the image are enhanced by fusion with the original image or convolution enhancement. The enhanced results are smoothed to suppress noise interference, thereby obtaining an enhanced visual image with clearer contours and more prominent structure.

[0021] The enhanced visual image is input into the feature extraction network, and its shallow semantic features such as edges and textures are extracted through shallow convolution layers. The high-resolution features are extracted in combination with the multi-scale structure to retain the spatial detail information of the image. The void rate is obtained according to a preset or adaptive method, and the corresponding void convolution kernel is generated accordingly. The contextual information of the image is extracted under different receptive fields through the void convolution operation, and the multi-channel features output by the void convolution are spliced and fused with the original features in the channel dimension to form a visual convolution feature containing multi-scale and multi-semantic information.

[0022] The extracted shallow semantic features, high-resolution features and visual convolution features are aligned according to the corresponding spatial dimensions, and the feature sizes are made consistent through upsampling or downsampling operations. Features at different levels are fused using skip connections, that is, shallow semantic features and high-level visual convolution features are spliced or weightedly superimposed in the channel dimension, thereby achieving complementary enhancement of low-level detail information and high-level semantic information, and obtaining semantic visual information containing rich texture structure and contextual semantics, providing multi-dimensional support for subsequent target detection or image understanding tasks.

[0023] Semantic segmentation of road visual images improves the clarity of edges and structural details while correcting image distortion, and fully extracts multi-level, multi-scale image feature information. By combining shallow semantic features, high-resolution details, and contextual features enhanced by dilated convolutions, and employing a skip connection fusion mechanism, it effectively achieves the coordinated expression of local details and global semantics, significantly improving segmentation accuracy and robustness. This provides more accurate and stable semantic visual information support for tasks such as object recognition, path extraction, and environmental perception in road scenes.

[0024] S2. Extracting weak geometric features of the road visual image, and constructing a scene semantic graph based on the semantic visual information and the weak geometric features.

[0025] In this embodiment of the present invention, to enhance the robustness of semantic segmentation in environments with blurred lane lines, occlusion, or unstructured roads, the collected road visual images are subjected to gradient enhancement and edge-preserving filtering. Convolutional gradient operators (such as Sobel and Scharr) are used to extract weak geometric features such as low-contrast edges, material change boundaries, and non-significant texture orientations. A geometric guidance graph is introduced to define spatial relationships in the scene by connecting paths. Based on semantic category labels and weak geometric constraints, a structurally complete and contextually consistent scene semantic graph is constructed, which serves as the basic input for downstream trajectory planning and path decision modules.

[0026] In detail, the extracting of weak geometric features of the road visual image includes: Converting the road visual image into a visual grayscale image; performing noise suppression on the visual grayscale image to obtain a smooth grayscale image; Calculating the horizontal gradient and the vertical gradient of the smoothed grayscale image, and generating a gradient magnitude map according to the horizontal gradient and the vertical gradient; Performing edge detection on the gradient magnitude map to obtain a weak geometric response map; Performing convolution coding on the weak geometric response map to obtain a geometric feature map code; performing mean pooling on the geometric feature map encoding and extracting a channel summary of the pooled geometric feature map; generating attention weights based on the channel summaries; The attention weight is used to perform feature weighted enhancement on the geometric feature map encoding to obtain weak geometric features.

[0027] In detail, the road visual image is converted into a single-channel visual grayscale image through weighted averaging or brightness channel extraction to simplify the image structure and retain key brightness information. The grayscale image is smoothed using noise suppression algorithms such as Gaussian filtering, median filtering or bilateral filtering to effectively remove random noise and subtle interference in the image while keeping the basic structure of the edge area from being blurred, ultimately obtaining a smooth grayscale image with uniform texture and clear boundaries.

[0028] Apply gradient detection operators such as the Sobel operator or the Prewitt operator to the smooth grayscale image, calculate the grayscale change rate of the image along the horizontal direction (x-axis) and the vertical direction (y-axis), and obtain the horizontal gradient map and the vertical gradient map respectively. According to the gradient values in these two directions, the Euclidean norm is used to calculate the gradient amplitude of each pixel to generate the gradient amplitude map. The calculation formula is: in, represents the horizontal gradient, Represents the vertical gradient.

[0029] By analyzing the gradient direction, the non-maximum suppression method is used to refine the gradient amplitude map, retaining only the pixels with local maximum values in the gradient direction to extract clear edge lines. A lower edge detection threshold is set, and the suppression results are preliminarily screened, retaining only pixel areas with low amplitude but potential edge features, thereby generating a weak geometric response map containing weak edge information.

[0030] The weak geometric response map is used as input and fed into a multi-layer convolutional neural network. The geometric structural features in the image are extracted through a series of convolutional layers and activation functions. The convolution kernel automatically learns the edge patterns and spatial relationships in the image, and gradually encodes the local geometric information into a high-dimensional feature representation. After convolution, pooling and nonlinear transformation, it finally generates a geometric feature map encoding with rich geometric expression capabilities, providing an effective feature basis for subsequent image understanding and analysis tasks.

[0031] The geometric feature map encoding is subjected to mean pooling in the spatial dimension to compress the spatial information of each channel into a single mean value, thereby obtaining a channel summary vector reflecting the overall response strength of each channel. The channel summary is used to generate attention weights through a fully connected layer or a lightweight neural network. These weights are used to represent the importance of each channel. The channels of the geometric feature map are weighted through the attention mechanism to highlight the key channel features.

[0032] The generated attention weights are multiplied element-by-element with the geometric feature map encoding according to the channel dimension, and the importance of each channel feature is dynamically adjusted, thereby amplifying the response signal of the key channel and suppressing irrelevant or redundant information, enhancing the more discernible part of the target edge and structure in the geometric feature map, and finally outputting the enhanced weak geometric features, providing more prominent and accurate feature expression for subsequent feature fusion and task inference.

[0033] In detail, constructing a scene semantic graph based on the semantic visual information and the weak geometric features includes: Acquire semantic entity types, and generate a semantic annotation graph according to the semantic visual information and the corresponding semantic entity types; generating a geometric guidance map according to the weak geometric features; generating a guidance weight map according to the semantic annotation map and the geometric guidance map; Performing structural guidance adjustment on the semantic annotation graph according to the guidance weight graph to obtain an updated semantic graph; Deconvolution is performed on the updated semantic graph to obtain a semantic structure graph; Using the connection paths in the geometric guidance graph as scene edges; Using the semantic nodes in the semantic structure graph as scene nodes; A scene semantic graph is generated according to the scene edges and the scene nodes.

[0034] Specifically, the semantic entity type categories in the road scene, such as lane lines, pedestrians, vehicles, etc., are obtained, the previously fused semantic visual information is associated with the corresponding semantic entity types, and each pixel point is mapped to the corresponding semantic category through pixel-level classification or segmentation algorithm. Finally, a semantic annotation map is generated to accurately display the spatial distribution and category information of each semantic entity in the image.

[0035] Based on the weak geometric features, a geometric guidance map is generated through methods such as convolution or interpolation to reflect the geometric structure and edge distribution characteristics in the image. The semantic annotation map is fused with the geometric guidance map, usually through weighted superposition or multi-channel splicing, and combined with an attention mechanism or a fusion network to calculate the guidance weight map, which effectively combines the advantages of semantic information and geometric features.

[0036] The updated semantic map is input into the deconvolution (transposed convolution) layer, which upsamples the semantic map to gradually restore the spatial resolution and refine the image structure information. The learnable convolution kernel is used to reconstruct and interpolate the features to enhance the continuity and clarity of the semantic boundaries, and finally outputs a high-resolution and structurally complete semantic structure map.

[0037] The paths representing connectivity in the geometric guidance graph are extracted and defined as scene edges, reflecting the geometric relationship between regions in space. The key semantic points identified in the semantic structure graph are used as scene nodes, representing important semantic entities in the scene. A graph structure is constructed based on these semantic nodes and semantic edges, and a complete scene semantic graph is formed through the connection relationship between nodes and edges. It comprehensively expresses the semantic entities in the scene and their geometric associations, providing structured information support for semantic reasoning and intelligent decision-making in complex environments.

[0038] Sophisticated gradient calculation and convolutional coding ensure accurate capture of geometric structural features, and the attention mechanism strengthens key channels, improving the pertinence and effectiveness of feature expression. Geometrically guided semantic annotation graph adjustment and deconvolution recovery enhance the spatial consistency and detail integrity of the semantic structure, and fuse geometrically guided paths and semantic nodes into a scene semantic graph, realizing the organic combination of semantic information and geometric relationships. This is conducive to improving the accuracy and robustness of target recognition, path planning, and scene understanding in road environments, and provides solid technical support for intelligent driving and related applications.

[0039] S3. Dynamically identify the central axis of the passable area of the scene semantic graph, and generate a plurality of candidate driving trajectories based on the identified central axis.

[0040] In an embodiment of the present invention, a dynamic main direction clustering algorithm based on geometric morphology and spatial consistency is adopted to adaptively extract the central axis of each labeled sub-region, expand the bandwidth around the central axis, and dynamically generate multiple candidate driving trajectory sets that meet the requirements of smoothness and controllability through curve fitting or trajectory sampling.

[0041] In detail, the dynamically identifying the central axis of the traversable area of the scene semantic graph includes: Identifying and labeling traversable areas on the scene semantic graph to obtain labeled areas; Counting the gradient directions of the pixels in the marked area, and generating a gradient angle map according to the gradient directions; Mapping the gradient direction to a preset unit circle coordinate according to the corresponding angle according to the gradient angle map to obtain a plurality of initial direction vectors; Clustering the initial direction vectors and calculating an average value of each cluster of initial direction vectors obtained by clustering; generating a main direction vector according to the average value; Determining whether the main direction vector corresponds to multiple initial direction vector clusters; If the main direction vector corresponds to multiple initial direction vector clusters, the marked area is divided according to the main direction vector to obtain multiple marked sub-areas; Performing binary refinement on each of the marked sub-regions to obtain a refined sub-region; Extracting skeleton lines of the refined sub-region, and selecting skeleton lines consistent with the main direction vector as candidate center lines; Sampling the candidate centerline to obtain a discrete point sequence; Performing curve fitting on the discrete point sequence to obtain a central axis; If the main direction vector corresponds to a unique initial direction vector cluster, the main direction vector is used as the central axis.

[0042] Specifically, based on the semantic labels of each region in the semantic map (such as roads, sidewalks, grass, buildings, etc.), image segmentation technology is used to extract regions with traversable attributes, such as roads and sidewalks. Morphological processing (such as dilation and erosion) is applied to remove noise and fill gaps to form continuous traversable areas. The contours of connected traversable areas are further extracted based on connectivity analysis and edge detection algorithms. Finally, these areas are labeled to indicate their traversability, providing a basis for subsequent path planning or navigation.

[0043] In the identified passable area, a gradient calculation operator (such as the Sobel operator) is applied to each pixel to extract its horizontal and vertical gradient components. The gradient direction angle of each pixel is calculated based on these gradient components. The inverse tangent function is usually used to obtain the accurate direction value. The gradient direction angle corresponding to each pixel is mapped to the pixel intensity or color value in the image to generate a gradient angle map, which is used to represent the local direction information of each pixel in the passable area.

[0044] According to the gradient direction angle of each pixel in the gradient angle map, the angle value is normalized to the angle range of the unit circle (such as 0°-360° or - arrive ), use trigonometric functions to map each angle to a direction vector on the unit circle, and generate the corresponding unit vector form for each angle ( ), the gradient direction of each pixel in the entire passable area corresponds to an initial direction vector, forming a vector set representing the local direction feature.

[0045] Using a clustering algorithm suitable for angular data (such as K-means, Mean-Shift, or clustering methods based on directional cosine similarity), vectors with similar directions are grouped into the same cluster. After clustering, all vectors within each directional vector cluster are averaged. This is typically done by averaging the horizontal and vertical components of each vector to calculate the cluster's average directional vector. These average vectors are then normalized to unit length to obtain the principal directional vector representing the dominant motion or structural trend in the region.

[0046] To determine whether the main direction vector corresponds to multiple initial direction vector clusters, we can analyze the number of clusters in the clustering results and the direction difference of each cluster: If the main direction vector corresponds to a unique initial direction vector cluster, the main direction vector is used as the central axis; if there are multiple main direction vectors (that is, the number of direction clusters is greater than 1 and the directions are significantly different), it means that the passable area contains multiple significant direction components. According to the direction of each main direction vector, the initial direction vector of each pixel is matched with the main direction vector for similarity (such as the maximum cosine similarity), and it is classified into the main direction category closest to it, thereby dividing the entire passable area into several labeled sub-areas with consistent directions.

[0047] Each annotated sub-region is binarized, with the pixels inside the region set as foreground (value 1) and the rest as background (value 0). An image thinning algorithm (such as Zhang-Suen or morphology-based thinning method) is then applied to the binary image to obtain the thinned sub-region. A skeleton extraction algorithm (such as distance transform-based skeleton extraction or MedialAxis transform) is used to extract the skeleton line of each thinned sub-region. By calculating the angle between the local direction of the skeleton line and the corresponding main direction vector, skeleton line segments with consistent or similar directions are screened out as candidate centerlines.

[0048] The selected candidate center lines are sampled at certain intervals, and a sequence of discrete points distributed along the line is extracted to simplify subsequent calculations. These discrete points are smoothly fitted using curve fitting methods (such as polynomial fitting, B-spline or spline interpolation) to generate a continuous, smooth curve that accurately reflects the shape of the original center line. The fitted curve is the extracted central axis, which can be used to represent the main traffic path or direction structure of the marked sub-area.

[0049] In an embodiment of the present invention, generating a plurality of candidate driving trajectories based on the identified central axis includes: Obtaining an offset range, and generating a plurality of driving trajectories to be selected according to the offset range and the central axis; calculating a curvature change of the driving trajectory to be selected, and determining whether the curvature change is smooth; If the curvature change is not smooth, deleting the driving trajectory to be selected corresponding to the non-smooth curvature change; If the curvature change is smooth, retain the driving trajectory to be selected corresponding to the non-smooth curvature change; The reserved driving trajectory to be selected is used as a candidate driving trajectory.

[0050] Specifically, the feasible offset range of the central axis is determined based on information such as road width, vehicle size, and traffic rules. It is usually set as a strip area within a certain distance to the left and right of the central axis. Along the normal direction of the central axis, several parallel curves are generated at fixed intervals within the offset range. Each curve represents a possible driving trajectory. These trajectories provide a variety of optional path solutions while ensuring that they do not exceed the scope of the marked sub-area.

[0051] The curvature of each candidate driving trajectory is calculated. This is typically based on the geometric relationships of the trajectory point sequence. Differentiation or curve-fitting derivatives are used to calculate how the curvature changes with path length. The continuity and smoothness of the curvature change are analyzed. Trajectories exhibiting sudden changes, violent fluctuations, or high-frequency oscillations are considered to have non-smooth curvature. These non-smooth trajectories are eliminated to avoid compromising driving stability and safety. Trajectories exhibiting smooth, continuous curvature changes without sudden changes are considered to meet the smoothness requirements and are retained. All trajectories with smooth curvature are considered candidate driving trajectories, providing a foundation for subsequent path selection and optimization.

[0052] By dynamically identifying the central axis of the traversable area in the scene semantic map and generating smooth and reasonable candidate driving trajectories based on the axis, it can adapt to complex road conditions and changes in traffic direction in different scenarios and achieve accurate modeling of the main traffic trend; by judging and screening curvature changes, it effectively eliminates non-smooth and unsuitable trajectories, ensuring that the generated candidate driving trajectories are geometrically continuous and smooth in driving, improving the safety and comfort of path planning, and providing the autonomous driving system with a more reliable set of driving path candidates.

[0053] S4. Analyze the feasibility of the candidate driving trajectory to obtain a feasibility probability.

[0054] In an embodiment of the present invention, the candidate trajectories are projected into the scene obstacle occupancy map, and semantic legitimacy verification is performed in combination with traffic semantic elements (such as lane lines, road boundaries, and dynamic obstacles). The passability probability is calculated based on multi-dimensional indicators such as collision risk, traffic continuity, structural continuity, and semantic complexity, and is used for subsequent path decision sorting.

[0055] Specifically, analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability includes: Obtaining obstacle information, and generating a scene obstacle occupancy map based on the obstacle information and the scene semantic map; Mapping the candidate driving trajectory into the scene obstacle occupancy map, and identifying the semantic category and obstacle occupancy status of each trajectory point of the candidate driving trajectory; Analyzing the collision risk of each candidate driving trajectory according to the obstacle occupancy state to obtain a collision probability; Analyzing the passage range of each candidate driving trajectory according to the semantic category to obtain passage continuity and structural continuity; Generating the semantic complexity of the area where each candidate driving trajectory is located according to the semantic category; The passability probability of each of the candidate driving trajectories is calculated according to the collision probability, the passability continuity, the structural continuity, and the semantic complexity.

[0056] Specifically, obstacle information is obtained, and the positions of static or dynamic obstacles are located and marked in the scene semantic map in combination with sensor data (such as lidar, camera or millimeter-wave radar). A scene obstacle occupancy map reflecting the spatial distribution of obstacles is generated, and the generated candidate driving trajectories are mapped point by point to the occupancy map. A semantic query is performed on each trajectory point to identify the semantic category of the location (such as road, lane line, non-traffic area, etc.) and whether it is occupied by an obstacle. This can effectively evaluate the passability and safety of the trajectory, and provide semantic and environmental constraints for subsequent path screening and decision-making.

[0057] Based on the obstacle occupancy status of each point on each candidate driving trajectory, the spatial overlap between the trajectory and obstacles is calculated. Incorporating information such as the obstacle type (static or dynamic), size, speed, and predicted position, a risk assessment is performed on points on the trajectory where collisions are likely. Based on the location distribution, distance, and dynamic trends of collision points, a probabilistic model (such as a Bayesian risk model or a distance-decay risk function) is used to calculate the collision probability for the entire trajectory, assigning a quantitative collision risk value to each candidate trajectory.

[0058] Candidate driving trajectories are mapped point by point onto the scene semantic graph, identifying the semantic category corresponding to each trajectory point (e.g., road center, lane line, sidewalk, building edge, etc.), and analyzing whether the trajectory points consistently reside within legal, traversable semantic areas. Traversable range analysis focuses on the spatial coherence of the trajectory and whether it crosses non-traversable areas. The coverage ratio of the trajectory in semantically consistent areas is calculated to assess traversability continuity. Structural continuity is assessed by combining semantic structure (e.g., road direction, lane width, and edge morphology) to determine whether the trajectory extends smoothly along the main road structure. Traversability continuity reflects whether the trajectory consistently stays within the permitted traversable area, while structural continuity reflects whether it conforms to road geometry and semantic trends.

[0059] Each candidate driving trajectory is sampled along the path and mapped into the scene semantic graph. The number and distribution of semantic categories within a certain range around the trajectory (e.g., the perception radius) are counted. If the trajectory contains multiple semantic types (such as lane lines, intersections, sidewalks, obstacle edges, etc.), with frequently changing boundaries and complex regional transitions, the region is considered to have high semantic complexity. If the semantic types are single, the boundaries are clear, and the structure is stable, the complexity is considered low. Based on factors such as semantic diversity, boundary density, and semantic transition frequency, a semantic complexity score is generated for each trajectory's region, providing a basis for evaluating safety and perceptual burden during route selection.

[0060] Collision probability reflects the risk of a trajectory colliding with obstacles in space, traffic continuity indicates whether the trajectory is always within the traversable area, structural continuity measures whether the trajectory is consistent with the road structure, and semantic complexity reflects the complexity and potential interference of the trajectory's environment. The traversable probability of each candidate driving trajectory is calculated using the following formula: in, represents the weight of the collision probability, represents the collision probability, represents the weight of traffic continuity, Indicates continuity of traffic. represents the weight of structural continuity, Indicates structural continuity, represents the weight of semantic complexity, Indicates semantic complexity, represents the passable probability.

[0061] The larger the collision probability value, the higher the risk; the higher the values of traffic continuity and structural continuity, the better the continuity; the lower the semantic complexity value, the simpler the environment; the higher the final weighted passability probability, the better the candidate driving trajectory in terms of safety, coherence, and environmental adaptability, and the more suitable it is as a candidate option for the final driving path.

[0062] By integrating obstacle information with scene semantic understanding, the system comprehensively evaluates candidate driving trajectories from multiple dimensions. This not only considers the collision risk between the trajectory and obstacles, but also integrates traffic continuity and road structural consistency to ensure that the path is spatially reasonable and structurally compliant with road rules. It also incorporates a semantic complexity metric to effectively identify high-risk areas or areas with high perceptual burden. Ultimately, through the fusion of these multiple factors, the system calculates the probability of traversability, providing a quantitative basis for path selection and significantly improving the safety, stability, and decision-making reliability of autonomous driving systems in complex environments.

[0063] S5. Filter the candidate driving trajectories for an optimal trajectory according to the passable probability to obtain a target driving trajectory of the target vehicle.

[0064] In an embodiment of the present invention, after obtaining the passable probabilities of candidate driving trajectories, all trajectories are dynamically sorted based on the passable probabilities. Combined with the driving intention of the target vehicle, semantic intention matching is introduced to remove redundancies from the retained trajectory set and generate the optimal target driving trajectory.

[0065] In detail, the optimal trajectory screening of the candidate driving trajectories according to the passable probability to obtain the target driving trajectory of the target vehicle includes: Determining whether the passability probability is greater than a preset score threshold; If the passable probability is less than or equal to the score threshold, the candidate driving trajectory corresponding to the passable probability less than or equal to the score threshold is deleted; If the passable probability is greater than the score threshold, then adding the candidate driving trajectory corresponding to the passable probability greater than the score threshold to the reserved trajectory set; Acquiring a navigation target of the target vehicle and generating a driving intention according to the navigation target; Matching the candidate driving trajectories in the retained trajectory set with the driving intention one by one to obtain an intention matching degree; The intention matching degrees are sorted in descending order, and the candidate driving trajectory corresponding to the highest intention matching degree after sorting is used as the target driving trajectory.

[0066] Specifically, the calculated passability probability of each candidate driving trajectory is compared with a score threshold. If the passability probability of a candidate driving trajectory is less than or equal to the score threshold, it indicates that there are significant risks in terms of safety, continuity, or environmental adaptability and is not suitable as a driving path. Therefore, this candidate driving trajectory is removed from the candidate set. If the passability probability of a candidate driving trajectory is higher than the score threshold, it indicates that it has high passability reliability. Then, the candidate driving trajectory is added to the retained trajectory set, effectively screening out unsafe or suboptimal paths, ensuring that the trajectory that finally enters the decision module has high driving feasibility and stability.

[0067] Obtain the target vehicle's navigation goal—the vehicle's current destination or the next navigation instruction (e.g., turn left, go straight, turn right, etc.). Combining the vehicle's current position, global path planning results, and road structure information, the system analyzes the target's positional relationship within the current road network and traffic requirements. Based on the navigation goal, it generates corresponding driving intentions, such as turning, changing lanes, staying straight, or slowing down to stop. This provides directional guidance for subsequent local path selection and behavioral decisions, ensuring safe and efficient vehicle travel in the correct direction.

[0068] The system calculates the spatial distance between the endpoint of each trajectory and the navigation target position, the consistency of the trajectory's overall direction with the driving intention direction, and whether the change in trajectory curvature meets the intention requirements. These factors are combined to obtain a quantitative intention matching score, which is used to evaluate the consistency between the trajectory and the intention. All trajectories are sorted from high to low according to the matching degree, and the trajectory that best meets the driving intention is screened out. The candidate driving trajectory with the highest matching degree is selected as the target driving trajectory, which is used to guide the vehicle's actual driving path and realize intention-driven precise path planning.

[0069] Through multi-level screening and matching, the safety and intelligence of route selection are effectively improved. High-risk or unreasonable trajectories are eliminated using passability probabilities, ensuring that the candidate path set has basic passability and stability. A clear driving intent is generated based on the vehicle's navigation goals, and retained trajectories are carefully matched based on the intent, ensuring that the final target driving trajectory is not only safe and reliable, but also accurately responds to driving needs and changes in the road environment. By comprehensively considering environmental risks and behavioral goals, the accuracy and dynamic adaptability of path planning are achieved, greatly improving the driving efficiency and safety of the autonomous driving system.

[0070] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0071] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the server side of a method for determining an intelligent driving trajectory based on a visual model.

[0072] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a method for determining an intelligent driving trajectory based on a visual model.

[0073] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed: Acquire a road visual image in a preset direction of the target vehicle, perform semantic segmentation on the road visual image, and obtain a plurality of semantic visual information; Extracting weak geometric features of the road visual image, and constructing a scene semantic graph based on the semantic visual information and the weak geometric features; Dynamically identifying a central axis of a traversable area of the scene semantic graph, and generating a plurality of candidate driving trajectories based on the identified central axis; Analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability; The candidate driving trajectories are screened for optimal trajectories according to the passable probabilities to obtain a target driving trajectory of the target vehicle.

[0074] In the several embodiments provided by the present invention, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative.

[0075] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0076] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0077] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0078] In some implementations of this embodiment, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0079] The readable storage medium of the present invention stores a computer program, which, when executed by a processor of an electronic device, can implement: Acquire a road visual image in a preset direction of the target vehicle, perform semantic segmentation on the road visual image, and obtain a plurality of semantic visual information; Extracting weak geometric features of the road visual image, and constructing a scene semantic graph based on the semantic visual information and the weak geometric features; Dynamically identifying a central axis of a traversable area of the scene semantic graph, and generating a plurality of candidate driving trajectories based on the identified central axis; Analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability; The candidate driving trajectories are screened for optimal trajectories according to the passable probabilities to obtain a target driving trajectory of the target vehicle.

[0080] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0081] The computer-readable storage medium may also store at least one computer-executable program / instruction, such as a computer-readable instruction. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), a hard disk, a flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above may be performed.

[0082] In addition, the computer device may also include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (eg, keyboard, mouse, speaker, etc.).

[0083] The processor can communicate with external devices via an I / O bus via a wired or wireless network.

[0084] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product / computer program product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0085] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0086] Those skilled in the art will clearly understand that for the sake of convenience and brevity in description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0087] In the embodiments provided in the present disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the systems, methods, and computer program products according to the various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the above-mentioned module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0088] It should be noted that, in this disclosure, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element limited by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0089] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for determining intelligent driving trajectory based on a visual model, characterized in that: The method comprises: Acquire a road visual image in a preset direction of the target vehicle, perform semantic segmentation on the road visual image, and obtain a plurality of semantic visual information; Extracting weak geometric features of the road visual image, and constructing a scene semantic graph based on the semantic visual information and the weak geometric features; Dynamically identifying a central axis of a traversable area of the scene semantic graph, and generating a plurality of candidate driving trajectories based on the identified central axis; Analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability; The candidate driving trajectories are screened for an optimal trajectory according to the passable probability to obtain a target driving trajectory of the target vehicle.

2. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: The semantic segmentation of the road visual image is performed to obtain a plurality of semantic visual information, including: Performing distortion correction on the road visual image to obtain a corrected visual image; performing edge enhancement processing on the corrected visual image to obtain an enhanced visual image; extracting shallow semantic features and high-resolution features of the enhanced visual image; Obtaining a dilation rate and a dilation convolution kernel corresponding to the dilation rate, and performing channel splicing on the enhanced visual image using the dilation convolution kernel to obtain a visual convolution feature; The shallow semantic features, the high-resolution features and the visual convolutional features are skip-connected and fused to obtain semantic visual information.

3. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: The extracting of weak geometric features of the road visual image includes: Converting the road visual image into a visual grayscale image; performing noise suppression on the visual grayscale image to obtain a smooth grayscale image; Calculating the horizontal gradient and the vertical gradient of the smoothed grayscale image, and generating a gradient magnitude map according to the horizontal gradient and the vertical gradient; Performing edge detection on the gradient magnitude map to obtain a weak geometric response map; Performing convolution coding on the weak geometric response map to obtain a geometric feature map code; performing mean pooling on the geometric feature map encoding and extracting a channel summary of the pooled geometric feature map; generating attention weights based on the channel summaries; The attention weight is used to perform feature weighted enhancement on the geometric feature map encoding to obtain weak geometric features.

4. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: The constructing of a scene semantic graph according to the semantic visual information and the weak geometric features includes: Acquire semantic entity types, and generate a semantic annotation graph according to the semantic visual information and the corresponding semantic entity types; generating a geometric guidance map according to the weak geometric features; generating a guidance weight map according to the semantic annotation map and the geometric guidance map; Performing structural guidance adjustment on the semantic annotation graph according to the guidance weight graph to obtain an updated semantic graph; Deconvolution is performed on the updated semantic graph to obtain a semantic structure graph; Using the connection paths in the geometric guidance graph as scene edges; Using the semantic nodes in the semantic structure graph as scene nodes; A scene semantic graph is generated according to the scene edges and the scene nodes.

5. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: The dynamically identifying the central axis of the traversable area of the scene semantic graph includes: Identifying and labeling traversable areas on the scene semantic graph to obtain labeled areas; Counting the gradient directions of the pixels in the marked area, and generating a gradient angle map according to the gradient directions; Mapping the gradient direction to a preset unit circle coordinate according to the corresponding angle according to the gradient angle map to obtain a plurality of initial direction vectors; Clustering the initial direction vectors and calculating an average value of each cluster of initial direction vectors obtained by clustering; generating a main direction vector according to the average value; Determining whether the main direction vector corresponds to multiple initial direction vector clusters; If the main direction vector corresponds to multiple initial direction vector clusters, the marked area is divided according to the main direction vector to obtain multiple marked sub-areas; Performing binary refinement on each of the marked sub-regions to obtain a refined sub-region; Extracting skeleton lines of the refined sub-region, and selecting skeleton lines consistent with the main direction vector as candidate center lines; Sampling the candidate centerline to obtain a discrete point sequence; Performing curve fitting on the discrete point sequence to obtain a central axis; If the main direction vector corresponds to a unique initial direction vector cluster, the main direction vector is used as the central axis.

6. The method for determining an intelligent driving trajectory based on a visual model according to claim 5, wherein: The step of generating a plurality of candidate driving trajectories according to the identified central axis includes: Obtaining an offset range, and generating a plurality of driving trajectories to be selected according to the offset range and the central axis; calculating a curvature change of the driving trajectory to be selected, and determining whether the curvature change is smooth; If the curvature change is not smooth, deleting the driving trajectory to be selected corresponding to the non-smooth curvature change; If the curvature change is smooth, retain the driving trajectory to be selected corresponding to the non-smooth curvature change; The reserved driving trajectory to be selected is used as a candidate driving trajectory.

7. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: Analyzing the feasibility of the candidate driving trajectory to obtain a feasibility probability includes: Obtaining obstacle information, and generating a scene obstacle occupancy map based on the obstacle information and the scene semantic map; Mapping the candidate driving trajectory into the scene obstacle occupancy map, and identifying the semantic category and obstacle occupancy status of each trajectory point of the candidate driving trajectory; Analyzing the collision risk of each candidate driving trajectory according to the obstacle occupancy state to obtain a collision probability; Analyzing the passage range of each candidate driving trajectory according to the semantic category to obtain passage continuity and structural continuity; Generating the semantic complexity of the area where each candidate driving trajectory is located according to the semantic category; The passability probability of each of the candidate driving trajectories is calculated according to the collision probability, the passability continuity, the structural continuity, and the semantic complexity.

8. The method for determining an intelligent driving trajectory based on a visual model according to claim 1, wherein: The step of screening the candidate driving trajectories for an optimal trajectory according to the passable probability to obtain a target driving trajectory of the target vehicle includes: Determining whether the passability probability is greater than a preset score threshold; If the passable probability is less than or equal to the score threshold, the candidate driving trajectory corresponding to the passable probability less than or equal to the score threshold is deleted; If the passable probability is greater than the score threshold, then adding the candidate driving trajectory corresponding to the passable probability greater than the score threshold to the reserved trajectory set; Acquiring a navigation target of the target vehicle and generating a driving intention according to the navigation target; Matching the candidate driving trajectories in the retained trajectory set with the driving intention one by one to obtain an intention matching degree; The intention matching degrees are sorted in descending order, and the candidate driving trajectory corresponding to the highest intention matching degree after sorting is used as the target driving trajectory.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the intelligent driving trajectory determination method based on a visual model as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for determining an intelligent driving trajectory based on a visual model as described in any one of claims 1 to 8 is implemented.

Images