Intelligent driving task configuration method and electronic device

Abstract

This invention discloses an intelligent driving task configuration method and electronic device, relating to the field of intelligent driving technology. The method includes determining track line data and driving environment type based on spatial data of the driving channel ahead of the intelligent device on the driving route; determining track complexity representation information of the intelligent device based on the track line data; determining scene complexity representation information based on the driving environment type; determining driving track scene complexity representation information based on the track complexity representation information and scene complexity representation information; and adjusting intelligent driving task configuration information based on the driving track scene complexity representation information. This invention can address the inability of related technologies to meet the timeliness requirements of user intelligent driving systems by accurately identifying road scene complexity and dynamically adjusting intelligent driving task configuration information, effectively improving the timeliness of intelligent driving systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, and in particular to an intelligent driving task configuration method and electronic device. BACKGROUND

[0002] With the development of intelligent driving technology, the requirements of intelligent driving systems are becoming higher and higher. Related technologies cannot balance timeliness and intelligent driving task accuracy when configuring intelligent driving tasks, and cannot meet the real-time needs of users for intelligent driving systems. SUMMARY

[0003] The present application provides an intelligent driving task configuration method and electronic device, which accurately identifies road scene complexity, dynamically adjusts intelligent driving task configuration information, and effectively improves the timeliness of intelligent driving systems.

[0004] To solve the above technical problems, the present application provides the following technical solutions:

[0005] In one aspect, the present application provides an intelligent driving task configuration method, comprising:

[0006] Obtaining driving channel space data in front of the intelligent device on the driving route; determining track line data according to the driving channel space data, and determining track complexity representation information of the intelligent device according to the track line data; determining driving environment types according to the driving channel space data, and determining scene complexity representation information according to the driving environment types; determining driving track scene complexity representation information according to the track complexity representation information and the scene complexity representation information, and adjusting intelligent driving task configuration information of the intelligent device according to the driving track scene complexity representation information.

[0007] In another aspect, the present application provides an electronic device, comprising a memory and a processor, wherein the processor is used to execute the computer program stored in the memory to realize the steps of any of the above intelligent driving task configuration methods.

[0008] Finally, the present application also provides an intelligent driving vehicle, comprising at least an intelligent driving perception system and a plurality of image acquisition devices for acquiring road data and environmental data in front of the intelligent vehicle; the intelligent driving perception system is used to execute the computer program stored in the memory to realize the steps of any of the above intelligent driving task configuration methods.

[0009] The intelligent driving task configuration method provided by the present application has the advantages that the complexity of the driving track is quantified from two dimensions of the driving track itself and the environment where the driving track is located, the overall complexity of the scene where the intelligent device is located at the next moment is comprehensively calculated, the configuration information such as the algorithm, the hardware and software resources, and the load used at the next moment is adjusted according to the overall complexity of the scene of the intelligent device, the intelligent driving task configuration information of the intelligent device is matched with the scene complexity of the driving track of the intelligent device, the task resources are accurately matched according to the scene complexity, the calculation load is reduced and the resources are saved in a simple scene, the algorithm strategy is optimized and the response speed is improved in a complex scene, the timeliness and the adaptability of the intelligent driving system are effectively improved, the road scene complexity is accurately identified, the intelligent driving task configuration information is dynamically adjusted, and the real demand of the user for the timeliness of the intelligent driving system is met.

[0010] In addition, the present application also provides corresponding electronic devices and intelligent driving vehicles for the intelligent driving task configuration method, which further makes the method more practical, and the electronic devices and intelligent driving vehicles have corresponding advantages. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiments or the related art descriptions. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0012] Figure 1 The hardware composition framework diagram for the intelligent driving task configuration method provided by the present application.

[0013] Figure 2 The flowchart of the intelligent driving task configuration method provided by the present application.

[0014] Figure 3 The topological structure diagram of the lane line in the exemplary application scenario provided by the present application.

[0015] Figure 4 The topological structure diagram in front of the intelligent device in the exemplary application scenario provided by the present application.

[0016] Figure 5 The two-dimensional lane line diagram of the exemplary application scenario provided by the present application.

[0017] Figure 6 The lane line diagram in the BEV perspective of the exemplary application scenario provided by the present application.

[0018] Figure 7An other-lane complexity diagram in the off-ramp situation in an example application scenario provided by the present application.

[0019] Figure 8 An other-lane complexity diagram in the on-ramp situation in an example application scenario provided by the present application.

[0020] Figure 9 A structural diagram of an example implementation of the intelligent driving vehicle provided by the present application.

[0021] Figure 10 A flow diagram of another intelligent driving task configuration method provided by the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. In the specification and the above drawings, the terms "first", "second", "third", "fourth" and the like are used to distinguish different objects, and are not intended to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. The term "exemplary" means "serving as an example, embodiment or illustration". Any embodiment described as "exemplary" herein is not necessarily interpreted as superior or better than other embodiments.

[0023] With the rapid development of artificial intelligence technology, the automatic driving technology has also been developed accordingly. The automatic driving system architecture has developed from the traditional modular architecture to the integrated perception and decision-making of end-to-end, and then to the world model, VLA (Vision-Language-Action Model) and fast-slow system based on large models. Although the model capability has been greatly improved, with the increase of model complexity, the model inference timeliness restricts the system deployment.

[0024] Different road scenarios (such as highways going straight, highways with ramps, closed urban roads, and urban traffic intersections) exhibit vastly different spatial and temporal complexities (e.g., peak and off-peak hours), leading to significant variations in the required algorithmic analysis capabilities. Related technologies, considering deployment and maintenance costs, often employ a unified pipeline or model to handle all road scenarios. For example, one related technology uses 3D CNN (Convolutional Neural Network) + RNN (Recurrent Neural Network) to identify traffic scenarios. While it effectively utilizes spatiotemporal information, its inherent computational bottlenecks and sequential nature make it highly unsuitable for autonomous vehicle control tasks with extremely high real-time requirements. Furthermore, the extremely high computational complexity and latency resulting from dual temporal modeling, unlike 2D convolution processing of a single image, 3D convolution requires simultaneous convolution operations in both spatial (height and width) and temporal (continuous video frame depth) dimensions, leading to an explosive increase in computational load. RNNs (especially LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit)) have an inherent serial dependency when processing sequential data. This means that the computation of step t+1 cannot begin until step t+1 is completed. This severely limits the parallel computing capabilities of hardware (GPU (Graphics Processing Unit) / TPU (Tensor Processing Unit)), resulting in slow inference speeds. Although a "two-layer" RNN may sound simple, its latency accumulates when processing long sequences. Another related technique is graph-based hazard scene recognition, which abstracts traffic scenes into graph structures and uses driving operation information to assist in hazard level identification. However, the larger the graph size (i.e., the number of nodes and edges), the longer the inference time of the GNN (Graph Neural Network). On open roads, the graph is small, and inference is fast, but at complex intersections, the graph becomes very large, and the inference latency increases significantly, demonstrating a real-time bottleneck in graph construction. Another related technology for traffic scene recognition involves: dividing LiDAR point clouds into planar point clouds (which may correspond to the ground, walls, etc.) and curved point clouds (which may correspond to vehicles, pedestrians, trees, etc.), and then identifying two-dimensional traffic information (such as lane lines and road boundaries) and three-dimensional traffic scene information (such as vehicles and pedestrians) for different types of point cloud data, and finally fusing the two-dimensional and three-dimensional data.Point cloud segmentation preprocessing incurs huge computational overhead. Dividing each frame of massive raw LiDAR point cloud into "planar" and "curved" planes requires running geometric feature extraction algorithms (such as those based on normal vector estimation and curvature calculation) or segmentation models (such as semantic segmentation networks based on deep learning). Such computationally intensive tasks also suffer from poor timeliness.

[0025] Therefore, the aforementioned technologies use a uniform task configuration to process all road scenarios, failing to consider the varying complexity of different scenarios. This leads to resource waste in simple scenarios and delayed responses in complex scenarios, making it impossible to balance timeliness and task accuracy, and thus failing to meet users' actual needs for intelligent driving systems. Another related technology is an adaptive object detection method based on scenario complexity pre-classification. This method uses a pre-classification network to determine scenario complexity and then adaptively selects object detection networks of different sizes to achieve dynamic model selection. However, lightweight pre-classification networks are prone to scenario classification errors, and the boundaries between "simple," "medium," and "complex" are inherently vague. For scenarios at the critical point, the pre-classification network may produce wavering judgments, leading to frequent switching of the backend model and introducing uncertainty, still failing to meet users' timeliness requirements for intelligent driving systems.

[0026] Therefore, this invention first rapidly assesses the complexity of the scene in which the intelligent device operates with fine granularity, and then selects different load models or algorithms for processing based on the scene complexity. This achieves dynamic adjustment of the autonomous driving system's algorithm and computational load, effectively improving the overall system's timeliness. The specific application environment architecture or specific hardware architecture upon which the execution of the intelligent driving task configuration method depends is described below. Figure 1 Examples of possible application scenarios related to the technical solutions of this invention are provided below:

[0027] Autonomous vehicles are equipped with at least three front-view image acquisition devices (such as a front left camera, a front center camera, and a front right camera) to collect multi-view image data. The multi-view image data is stitched together, and image features are extracted. These image features, along with their timestamps, are placed into a front-view image feature queue, while navigation data is acquired simultaneously. Autonomous vehicles navigate mixed urban and highway roads, including various scenarios such as urban intersections, highway straightaways, highway ramps, and mountain curves. The intelligent driving task configuration needs to be dynamically adjusted according to the complexity of different scenarios.

[0028] In urban road areas, navigation data is complete, and image data and navigation data are used as the driving lane spatial data. In mountainous road areas, navigation data is missing, and stitched image data is used as the driving lane spatial data. In urban road areas, navigation data undergoes color layering, binarization, and edge processing to extract lane line topology information and determine the target lane line and adjacent lane lines. In mountainous road areas, target image features are queried from the front view image feature queue to identify two-dimensional lane lines, which are then converted into lane line data under BEV (bird's-eye view) using image projection transformation methods. For urban road scenarios, lane lines are fitted using a third-order curve to determine that the intelligent vehicle is ahead of a U-shaped curve, and the lane complexity representation information for the autonomous vehicle is calculated to be 0.6. Based on the driving lane spatial data, the lighting type is identified as daytime backlight and the weather type as sunny, with a scene complexity representation information of 0.75; the overall driving lane scene complexity representation information is 0.45. For a straight highway scenario, the target lane is straight, with a lane complexity of 0.3; the lighting type is daytime with direct sunlight, and the weather type is sunny, resulting in a scene complexity of 0.25; the overall complexity is 0.075. For a highway ramp scenario (including exit ramps), the target lane is an S-shaped curve, with a lane complexity of 0.8; the lighting type is evening with direct sunlight, and the weather type is cloudy, resulting in a scene complexity of 0.5; the overall complexity is 0.4. For a mountain curve scenario, the target lane is a U-shaped curve, with a lane complexity of 0.7; the lighting type is daytime with direct sunlight, and the weather type is rainy, resulting in a scene complexity of 0.8; the overall complexity is 0.56.

[0029] When the complexity of the driving trajectory scene corresponding to an urban intersection is determined to be 0.45, the sensor sampling frequency in the intelligent driving task configuration information can be increased to 15 frames / second, a high-precision steering decision algorithm can be enabled, and the driving speed can be reduced to 30 km / h. When the complexity of the driving trajectory scene corresponding to a highway straight road is determined to be 0.075, the sensor sampling frequency in the intelligent driving task configuration information can be reduced to 8 frames / second, a lightweight decision algorithm can be adopted, and the driving speed can be maintained at 100 km / h. When the complexity of the driving trajectory scene corresponding to a highway ramp is determined to be 0.4, a lane-changing path can be planned 500 meters in advance, the driving speed can be reduced to 60 km / h, and the monitoring accuracy of adjacent lanes can be enhanced. When the complexity of the driving trajectory scene corresponding to a mountain curve is determined to be 0.56, the image enhancement algorithm strength can be increased, the braking distance can be extended, the driving speed can be controlled at 40 km / h, and the anti-skid steering strategy can be optimized.

[0030] As can be seen from the above, the autonomous vehicle in this embodiment, during driving on mixed road sections, saves computing resources and improves driving efficiency in simple scenarios by dynamically adjusting the intelligent driving task configuration, while ensuring driving safety and stability in complex scenarios, effectively avoiding problems such as decision delays and loss of steering control. It should be noted that the above application scenarios are only shown to facilitate understanding of the ideas and principles of the present invention, and the implementation of the present invention is not limited in any way. On the contrary, the implementation of the present invention can be applied to any applicable scenario. Various non-limiting embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Please see first. Figure 2 According to the intelligent driving task configuration method provided by the present invention, it can be implemented as a computer program product, for example, installed and running in the intelligent driving perception system of an intelligent device, for dynamically adjusting the intelligent driving task configuration information of the intelligent device during driving. In some embodiments of the method, the following steps may be included:

[0032] S201: Obtain spatial data of the driving lane ahead of the intelligent device on the driving route.

[0033] In this context, "intelligent device" refers to a carrier with intelligent driving capabilities, including but not limited to intelligent driving vehicles, autonomous robots, and other devices capable of autonomous driving on roads. This invention also applies to devices that fly intelligently on waterways, such as drones, and devices that move intelligently on water. Accordingly, the travel trajectory of an intelligent device can include roads, waterways, and air routes / waterways. Travel channel spatial data describes the travel trajectory ahead of the intelligent device and its surrounding environment. This data can include image data and navigation data and serves as the foundational data source for subsequently determining the trajectory line and travel environment type. A travel channel is a defined space with width or range. The travel channel spatial data at least includes data within that channel that can determine the travel trajectory line and surrounding environmental information (such as lighting and weather), enabling a comprehensive analysis of the travel trajectory scenario and a fine-grained evaluation of the trajectory scenario from multiple aspects, including environment, lighting, track topology, and scene.

[0034] S202: Determine the track line data based on the driving channel space data, and determine the track complexity representation information of the intelligent device based on the track line data.

[0035] Track data refers to information reflecting the track topology, including track regions, track lines, the track where the smart device is located, and its adjacent track lines, used to analyze track complexity. Track complexity representation information is a parameter that quantifies the complexity of the track itself and its adjacent tracks, and can include the complexity of the target track and the complexity of adjacent tracks.

[0036] S203: Determine the driving environment type based on the driving channel space data, and determine the scene complexity representation information based on the driving environment type.

[0037] Among them, driving environment type refers to the environmental category based on factors such as lighting and weather, such as daytime with direct sunlight, daytime with backlight, nighttime with direct sunlight, nighttime with backlight, sunny day, rainy day, foggy day, etc. Any classification and recognition network model with the ability to identify lighting type and weather type can be used to determine the driving environment type. The scene complexity representation information is the environmental complexity parameter quantified based on the driving environment type.

[0038] S204: Determine the driving trajectory scene complexity representation information based on the driving trajectory complexity representation information and the scene complexity representation information, and adjust the intelligent driving task configuration information of the intelligent device based on the driving trajectory scene complexity representation information.

[0039] Among them, the driving trajectory scenario complexity representation information is a parameter that comprehensively reflects the overall complexity of the trajectory scenario by combining trajectory complexity representation information and scenario complexity representation information. The intelligent driving task configuration information consists of the task parameter settings related to driving in the intelligent driving system, including sensor operating modes, algorithm types, decision response strategies, computational load, storage resources, and computing resources. This allows for adjustments to things like autonomous driving system algorithms and computational load based on scenario complexity, significantly improving overall system efficiency.

[0040] In the technical solution provided in this embodiment, the complexity of the driving channel space is quantified from two dimensions: the driving track itself and the environment in which the driving track is located. The overall complexity of the scenario in which the intelligent device will be located at the next moment is then considered. Based on the overall complexity of the scenario, the configuration information such as the algorithm, hardware and software resources, and load used at the next moment is adjusted. This ensures that the intelligent driving task configuration information of the intelligent device matches the complexity of the driving track scenario. Task resources are accurately matched according to scenario complexity, reducing computational load and saving resources in simple scenarios, while optimizing algorithm strategies and improving response speed in complex scenarios. This effectively improves the timeliness and adaptability of the intelligent driving system, thereby achieving the goal of accurately identifying road scenario complexity and dynamically adjusting intelligent driving task configuration information to meet users' real-world needs for the timeliness of the intelligent driving system.

[0041] Considering the current lane line extraction algorithms based on deep learning, such as the Transformer architecture, which has a low inference speed, CNN-based algorithms offer a significant speed improvement but suffer from the following problems: Lane line loss: Vehicle occlusion leads to incomplete lane line recognition; Accuracy issues: Lane lines are difficult to accurately identify when they are unclear or obscured by rain or snow; Viewpoint transformation issues: Lane lines are typically obtained in the image coordinate system using CNN classification algorithms. Current lane line recognition algorithms suffer from the following problems: Lane line loss due to vehicle occlusion; BEV algorithms based on Transformer suffer from computational time issues. Image point classification algorithms for lane line recognition do not directly reflect lane topology due to camera projection problems. Navigation data, such as navigation maps, can provide lane-level maps from a BEV perspective, and road-related information can be obtained through simple processing. This embodiment provides a method that uses both navigation data and image data as driving lane spatial data, which may include the following:

[0042] Acquire image data of the front of the smart device while it is driving; acquire navigation data of the smart device; if the navigation data is missing or the driving track in the navigation data is incomplete, then use the image data as the driving channel space data; if the navigation data exists and the driving track information in the navigation data is complete, then use the navigation data and image data together as the driving channel space data.

[0043] Image data refers to images of the road ahead and its surrounding environment captured by image acquisition devices such as cameras mounted on smart devices. These images can include single-view and multi-view stitched images. Navigation data provides driving-related information to the smart device's navigation system, including track topology, track distribution, and ramp locations. Missing track information means the navigation map data does not fully record current track information, such as the number and shape of tracks. Complete track information means the navigation data accurately records all the information the user needs, including the number of tracks, track topology, curve shapes, and ramp distribution.

[0044] In this embodiment, navigation data can provide a track-level map from the BEV's perspective, making it easy to obtain accurate driving tracks with minimal computation and enabling parallel processing. Furthermore, with continuous improvements and updates, the local completeness of the navigation data can meet users' needs for track-related information. Typically, navigation maps contain most road data, but in certain special cases, navigation may be missing or incomplete. Therefore, to analyze and obtain the track data of the intelligent device at the current and next moment, it can be processed separately. For example, when an intelligent driving vehicle is traveling on a highway, the navigation map data completely records the road track topology, curve distribution, and ramp location information. The vehicle collects real-time image data of the road ahead through a forward-facing camera, simultaneously acquiring navigation map data. Both types of data are used together as the driving channel spatial data, where the navigation map data is used to determine the basic track topology, and the image data is used to correct the real-time track status (such as positional deviations caused by lane line wear). When the vehicle travels to a newly built urban road, the navigation map data for this area has not yet been updated, and driving track information is missing. The vehicle automatically switches data sources, stitching together multi-view image data from the front-view cameras (such as the front left, front center, and front right cameras) to create the driving lane spatial data. Track lines and environmental information are then extracted using image recognition technology. This ensures the validity of the driving lane spatial data in both scenarios, providing a reliable foundation for subsequent complexity assessments and task configuration adjustments, and preventing system decision-making errors due to data gaps.

[0045] As shown above, this embodiment establishes a complementary acquisition mechanism for image data and navigation map data, dynamically selecting the data source based on the completeness of the navigation map data. When the navigation map data is complete, the advantages of both types of data are combined, namely the accurate topology of navigation and the real-time nature of image data; when the navigation map data is missing or incomplete, image data is used as a supplement. This ensures the integrity and reliability of the driving channel spatial data, providing high-quality data support for subsequent assessments of track complexity and scene complexity, and improving the adaptability of the intelligent driving system.

[0046] To quickly and accurately extract orbital data from navigation data and reduce the computational resources used in this process, the present invention also provides the following implementation methods, which may include:

[0047] The navigation data is processed into sub-images of different color layers; each sub-image of a color layer is binarized; each binarized image is processed using an edge algorithm to obtain track data, and track line topology information is determined based on the track data; according to the spatial location of the smart device, the target track line topology information ahead of the smart device is determined based on the track line topology information, which is used as track line data.

[0048] Among them, the aerial view mode is one of the navigation display modes, providing a vertical downward perspective of the road and clearly showing the topological structure of the road tracks. Color layering is a process of dividing navigation data into multiple color layers based on the color differences of different elements in the navigation (such as lane lines, road surface, and green belts). A sub-image refers to an image fragment corresponding to a single color layer obtained after color layering. Binarization is an image processing technique that converts sub-images into images containing only black and white, creating a sharp contrast between target elements (such as lane lines) and the background (such as road surface). Edge algorithms are used to extract edge features of target elements in the image; in this embodiment, the Sobel algorithm or the Canny algorithm is used. Track data is image data reflecting the edges and shapes of the tracks, extracted through edge algorithms. Track topology information is structured information describing the connection relationships and positional distribution between all tracks. Target track topology information is track topology information for a specific range ahead of the smart device and can be directly extracted from the track topology information.

[0049] In this embodiment, to minimize unnecessary calculations, the navigation map can be adjusted to BEV mode with the vehicle facing upwards, i.e., an aerial view, avoiding the calculation process required to change the perspective of lane lines due to different viewpoints. Typically, once the navigation software is set up, the road surface and lane line colors in the navigation map remain largely unchanged. Through color layering, edge extraction operators can be used to quickly obtain edge information for different color layer images. For example, when the navigation map can obtain the lane line topology information in front of the vehicle from a BEV perspective, layering is performed according to the color settings to obtain images of different color layers. These images are then binarized, and edge algorithms such as Sobel and Canny are used to obtain the required information such as lane areas, double yellow lines, and lane lines. Since different color areas correspond to different types of roads, a semantic-level lane line topology map can be further constructed, such as... Figure 3 As shown, to expedite the subsequent track data acquisition process and reduce resource requirements, further topology information ahead of the intelligent device can be obtained. This allows for the identification of the vehicle's location and lane on the navigation map. Figure 3 The lane topology diagram is cut off to extract the lane line topology diagram M meters in front of the vehicle, such as... Figure 4 As shown in the figure, the lane formed by the two thicker straight lines is the current lane.

[0050] As shown above, this embodiment uses a combination of color layering, binarization processing, and edge algorithms to quickly separate and extract track-related data, construct track topology information, and then combine this with the spatial location of the intelligent device to filter out the target track topology information. This achieves efficient and accurate extraction of track data, improving the efficiency of track complexity assessment. Simultaneously, the navigation map data based on the aerial view mode ensures the accuracy of the track topology information, enabling the system to quickly and accurately determine track type and curvature characteristics, providing a reliable basis for subsequent intelligent driving task configuration adjustments.

[0051] Considering that the field of view of images captured by a single camera is limited and it is difficult to fully cover the road scene ahead, and that the image features are not stored in a time-related manner, resulting in low efficiency in subsequent queries of image features at specific times, affecting the timeliness of track line data and environment type judgment, based on the above embodiments, the present invention also provides an exemplary implementation of acquiring image data of the front of a smart device, which may include the following: acquiring images corresponding to multiple perspectives of the front of the smart device and stitching the images together to form image data; extracting image features from the image data and setting acquisition time identifiers for the image features; and adding each image feature to a pre-constructed front view image feature queue.

[0052] In this context, "multiple viewpoint images" refers to road images captured from different angles by multiple front-facing cameras (such as front left, front center, and front right cameras) on a smart device. Image stitching refers to the process of combining images from multiple viewpoints into a single, large-field-of-view image using image fusion technology. Image features refer to key information in the image that reflects road and environmental characteristics, including lane line edges, color distribution, and light intensity. Acquisition timestamps are timestamps added to each image feature to mark the corresponding travel time. The front-view image feature queue is an ordered data structure storing front-view image features with timestamps, used for quick subsequent lookups of image features at specific times.

[0053] In this embodiment, to obtain more comprehensive track and scene information, images from the three front cameras in the surround-view camera can be stitched together. Features are then extracted from the stitched image, for example, using ResNet (network model name) or other commonly used image backbone networks to extract image feature backbones. The acquisition time is then used as a timestamp to add the image features to a queue for use by subsequent modules. For instance, an intelligent driving vehicle is equipped with three cameras—front left, front center, and front right—to capture road images from different angles ahead at a rate of 10 frames per second. An image stitching algorithm (such as SIFT feature matching) is used to stitch the images from the three cameras into a complete image with a large field of view, covering a 120° view in front of the vehicle. Feature extraction is performed on the stitched image to obtain lane edge features, road surface color features, and light intensity features, and a corresponding acquisition time identifier (e.g., "2025-10-01 14:30:01") is added to each feature. Image features with time stamps are added to the forward image feature queue in chronological order of acquisition, with a queue capacity of 100 entries, using a first-in, first-out (FIFO) storage strategy. When the vehicle travels to an area where navigation map data is missing, the system quickly queries the feature queue for image features at the current time and the most recent historical time to extract track data and determine the driving environment type, ensuring continuous and stable system operation.

[0054] As shown above, this embodiment expands the field of view of the image through multi-view image stitching, ensuring comprehensive capture of road and environmental information ahead. By adding time stamps and constructing a feature queue, it achieves orderly storage and rapid retrieval of image features. This improves the integrity and query efficiency of image data, providing support for quickly obtaining track line data and environmental information in scenarios without navigation map data, and ensuring the real-time response capability of the intelligent driving system.

[0055] Furthermore, for scenarios where there is no navigation data or the navigation data is incomplete, the process of acquiring track line data can be as follows: obtain target image features within the time range of the target's travel time from the front view image feature queue; identify two-dimensional track lines in the target image features and convert the two-dimensional track lines into track lines in a bird's-eye view to serve as track line data.

[0056] The target travel time range is a time interval preset by the user or system for querying image features; in this embodiment, it is set to 0.5 seconds before and after the current time of the smart device. The target image features are image features within the target travel time range in the front-view image feature queue. A two-dimensional track line refers to a track line identified in the image coordinate system, reflecting only the shape of the track in a plane and not considering spatial perspective differences, such as... Figure 5As shown. A bird's-eye view is a perspective taken from above, looking vertically down at the road. It provides a clear view of the track's topology and relative positions, such as... Figure 6 As shown.

[0057] In this embodiment, to minimize computational load, instead of extracting image features when analyzing trackline data, the camera image features from the target recognition process are added to a queue, and subsequently read from the queue. Time alignment requirements are not high here; the scene changes within a few seconds of the vehicle's movement, which is relatively minor. When the required navigation data is unavailable, the forward-view image feature queue is queried, and the image features of the stitched image closest to the current time are obtained based on the current timestamp. From these image features, 2D lane lines are identified. To improve the overall processing speed, a lightweight lane line recognition algorithm (such as ultra-fast-lane-detection) can be used. Due to the camera imaging principle, 2D lane lines do not represent the BEV perspective and need to be converted to BEV lane lines. Based on image projection transformation methods such as IPM (Inverse Perspective Mapping), the lane line results are converted to the BEV perspective using a camera projection transformation matrix. For example, when an autonomous vehicle travels on a rural road where navigation map data is missing, the system sets the target travel time range to 0.5 seconds before and after the current time (e.g., 14:35:20). It queries the front-view image feature queue for target image features within this time range, including lane edge features and road surface features from a 3-frame stitched image. Based on a lane detection model, it identifies the two-dimensional track line (i.e., the lane outline in the image coordinate system) from the target image features. Using the IPM method, it constructs a transformation matrix based on camera intrinsic and extrinsic parameters to convert the two-dimensional track line into a bird's-eye view track line, obtaining track line data that accurately reflects the road's track topology. Based on this track line data, the system determines that the current road is straight and has low track complexity, thus adjusting the task configuration to reduce computational load and improve driving efficiency.

[0058] As shown above, this embodiment ensures the timeliness of track line recognition by filtering image features within the target time range; it converts two-dimensional track lines into track lines from a bird's-eye view, correcting the perspective deviation of the planar image and accurately restoring the topological structure and relative position of the track. This significantly improves the accuracy of track line data acquisition in scenarios without navigation map data, providing a reliable basis for track complexity assessment and ensuring that the intelligent driving system can still accurately adjust task configuration in this scenario.

[0059] In the above embodiments, no limitation is made on how to determine the orbital complexity representation information. Based on the above embodiments, the present invention also provides an exemplary implementation of determining the orbital complexity representation information of a smart device based on orbital line data, which may include the following:

[0060] Based on the track data, determine the target track line and adjacent track lines of the target track where the intelligent device is located; based on the position information of each track position point on the same side of the target track line, perform curve fitting processing of no less than order three to obtain the track fitting curve; based on the extreme points and coefficients of the track fitting curve, determine the track type of the target track, and determine the target track complexity representation information based on the track type and the curvature information of the track fitting curve; based on the track type of the target track, the driving action of the intelligent device and adjacent track lines, determine the adjacent track complexity representation information in front of the intelligent device; based on the target track complexity representation information and the adjacent track complexity representation information, determine the track complexity representation information.

[0061] The target track line is the track currently being traveled by the smart device. Adjacent track lines are those directly adjacent to the target track line, including the left and right adjacent track lines. Track position points are coordinate points located on the track line, describing the track's shape. Curve fitting is a mathematical modeling process that approximates the distribution of multiple data points using a smooth curve. Considering the characteristics of different tracks, a curve fitting of at least order three is used to accurately capture the track's curvature. The track fitting curve is the curve obtained after curve fitting that reflects the actual shape of the track. Extreme points are points on the track fitting curve where the derivative is zero, used to determine the track's curvature type (e.g., single-inflection curve, multi-inflection curve). Track type is a category based on track shape, including straight tracks, type I curves (including multiple inflection points, such as S-shaped tracks), and type II curves (including one inflection point, such as U-shaped tracks). Curvature information is a parameter describing the curvature of the track fitting curve, including average curvature, maximum curvature, and standard deviation of curvature. The target track complexity is a parameter that quantifies the complexity of the track itself on which the intelligent device resides. Driving actions refer to the intelligent device's driving planning behavior, including leaving the target track or staying on it. Adjacent track complexity is a parameter that quantifies the complexity of adjacent track lines.

[0062] In this embodiment, considering the limited length range of local lane lines, a cubic curve fitting method is used, which can effectively handle S-shaped and U-shaped curves. The complexity of the track where the intelligent device is located is described in fine granularity from multiple dimensions. By normalizing the curvature and combining four indicators, the complexity of the track where the intelligent device is located is normalized. Intuitively, straight tracks have relatively low complexity, while curved tracks have relatively high complexity. For curves, such as S-shaped curves, at least two turns are required.

[0063] For example, during the operation of an intelligent driving vehicle, the target trajectory line (i.e., the current driving lane) and adjacent trajectory lines (i.e., the lanes to the left and right of the current driving lane) are determined using spatial data of the driving lane. The positional information of 100 trajectory points on the same side of the target trajectory line is extracted, and a third-order curve fitting process is used to obtain the trajectory fitting curve, which can represent... Let x and y be the position coordinates. Differentiating the trajectory fitting curve yields the derivative equation. Solving for the derivative reveals two extreme points, confirming the target track type as a first-type curve. The average curvature of the trajectory fitting curve is calculated to be 0.6, the maximum curvature to be 0.9, and the standard deviation of curvature to be 0.2. Combining this with the steering complexity representation value corresponding to the first-type curve, the target track complexity representation is determined to be 0.7. Given that the intelligent device's driving action is to remain on the target track and adjacent track lines do not include ramps, based on the target track type (e.g., first-type curve) and driving action, the adjacent track complexity representation is determined to be 0.2. Combining the target track complexity representation (0.7 in this embodiment) and the adjacent track complexity representation (0.2 in this embodiment), the track complexity representation is obtained as 0.45. This result accurately reflects the complexity of the current track and adjacent tracks, providing a precise basis for adjusting the sensor detection range and algorithm calculation accuracy.

[0064] As shown above, this embodiment accurately captures the track shape through third-order or higher curve fitting, subdivides track types and quantifies the complexity of the target track by combining extreme points and curvature information; simultaneously, it considers the correlation between the target track type, driving actions, and adjacent track lines to comprehensively evaluate the complexity of adjacent tracks, ultimately obtaining more accurate track complexity representation information. This provides a reliable basis for subsequent scenario complexity assessment, making the configuration of intelligent driving tasks more closely match actual track conditions, and improving driving safety and smoothness.

[0065] For example, this embodiment also provides a method for identifying the track type of the target track: if the coefficients corresponding to the highest order and the second highest order of the track fitting curve are both less than preset values, then the target track is a straight track; if the target track is not a straight track and the number of extreme points is not less than 2, then the target track is a first type of curve with multiple inflection points; if the target track is not a straight track and the number of extreme points is less than 2, then the target track is a second type of curve with one inflection point.

[0066] The coefficient corresponding to the highest order is the coefficient of the term with the highest degree in the orbital fitting curve equation, such as the fourth-order curve. The coefficient 'a' in the equation represents the coefficient of the term with the second-highest order in the trajectory fitting curve equation, such as the coefficient 'b' in the fourth-order curve mentioned above. x and y are the position coordinates, and a, b, c, d, and e are the coefficients of the trajectory fitting curve. The preset value is a threshold determined based on a large amount of road data statistics, used to determine whether the track is a straight road. In this embodiment, the preset value is set to 0.01 for example. The first type of curve refers to a curve with multiple inflection points (i.e., the number of extreme points). Two types of curves, such as S-curves, have two inflection points. The second type of curve refers to a curve with one inflection point (i.e., the number of extreme points). Two types of curves, such as U-turns and arc curves. For example, for the lane where the smart device is located, arbitrarily select the left or right lane line and perform a third-order curve fitting. The fitted curve can be represented as... The coefficients a and b are selected to be very small numbers close to 0; otherwise, the road is not a straight road, but is defined as a curve in this embodiment. For a curve, the extreme points of the track fitting curve are calculated; if there are 2 extreme points, it is an S-curve with 2 turns; if there is 1 extreme point, it is a U-curve with 1 turn. For example, when an intelligent driving vehicle is driving on a mountain road, after acquiring the trajectory point data of the target track line, a third-order curve fitting is performed to obtain the track fitting curve. The highest-order coefficient is 0.008, and the second-highest-order coefficient is 0.005, both greater than the preset value of 0.01, indicating that the trajectory is not a straight path. The derivative of the fitted curve is obtained... Setting the derivative to zero, solving the equation yields two extreme points: and Number of extreme points Two points indicate that the target track is a Type I curve (e.g., an S-shaped curve). Based on this track type, the system will subsequently focus on the frequent turning issues caused by multiple inflection points when calculating track complexity, adjusting the response speed and accuracy of steering decisions to ensure smooth steering of the vehicle when driving on an S-shaped curve and avoid the risk of rollover.

[0067] As shown above, this embodiment determines whether a road is a straight section by using the highest-order and second-highest-order coefficients of the trajectory fitting curve, and then further subdivides the curve type based on the number of extreme points, achieving accurate classification of track types. This provides a foundation for subsequent targeted calculations of the complexity of different types of tracks, enabling the intelligent driving system to adjust its driving strategy according to the different characteristics of straight sections, single-turn curves, and multi-turn curves, thereby improving the safety and stability of cornering.

[0068] For example, after determining the track type and the curvature information of the track fitting curve, the process of determining the target track complexity representation information may include: obtaining preset straight track complexity values ​​and preset curve complexity values; determining the overall curvature representation value, curvature variation representation value, and curve curvature value of the target track based on the curvature information of the track fitting curve; determining the curve quantization value based on the steering complexity representation value, overall curvature representation value, curvature variation representation value, and curve curvature value corresponding to the first type of curve and the second type of curve; determining the curve complexity representation value based on the preset straight track complexity value, the preset curve complexity value, and the curve quantization value; if the target track is a straight track, the target track complexity representation information is the preset straight track complexity value; if the target track is not a straight track, the target track complexity representation information is the curve complexity representation value; wherein, the steering complexity representation value corresponding to the first type of curve is greater than the steering complexity representation value corresponding to the second type of curve.

[0069] The preset straight-line complexity value is a pre-set complexity value corresponding to a straight road. It can be flexibly adjusted according to actual needs and determined based on the typical difficulty of driving on a straight road. In this embodiment, it can be set to 0.3 for example. The preset curve complexity value is a benchmark value used to calculate curve complexity. It can be flexibly adjusted according to actual needs, with different values ​​for different types of curves. For curves, if it is a first-type curve such as an S-curve, the preset curve complexity value can be 1.0; if it is a second-type curve such as a U-curve, the preset curve complexity value is 0.5. The overall curvature degree representation value is a parameter quantified by the average curvature of the track fitting curve, reflecting the overall curvature trend of the track. The curvature variation degree representation value is a parameter quantified by the standard deviation of the curvature of the track fitting curve, reflecting the fluctuation of the track curvature degree. The curve curvature value is the curvature data of each point on the track fitting curve, including key parameters such as the maximum curvature value. The steering complexity representation value is a quantified parameter set according to the curve type, used to reflect the impact of the number of turns on the complexity. The cornering quantification value is an intermediate quantification result of the cornering complexity, obtained by combining the steering complexity representation value, the overall curvature representation value, the curvature variation representation value, and the cornering curvature value. For example, the sum of the steering complexity representation value, the overall curvature representation value, the curvature variation representation value, and the cornering curvature value, divided by 4, can be used as the cornering quantification value. The cornering complexity representation value is the final quantized cornering complexity parameter. For example, the product of the preset cornering complexity value and the cornering quantization value, plus the sum of the preset straight-line complexity value, can be used as the cornering complexity representation value.

[0070] For example, the preset straight-line complexity value is 0.3, and the preset curve complexity value is 0.5. When an autonomous vehicle travels to a curved road, it is determined by track type to be a Type I curve (S-shaped), and its steering complexity value is set to 1.0. The average curvature of the fitted curve for this curve is calculated to be 0.7 (i.e., the overall curvature value), the standard deviation of curvature is 0.3 (i.e., the curvature variation value), and the maximum curvature is 0.9 (i.e., the curve curvature value). Based on preset weighting coefficients (e.g., steering frequency weight is 0.3, curve weight is 0.2, curve variation weight is 0.2, and curve curvature weight is 0.3), these weights can be adjusted according to the actual scenario. The calculated curve quantification value is: 1.0×0.3+0.7×0.2+0.3×0.2+0.9×0.3=0.3+0.14+0.06+0.27=0.77. Based on the preset curve complexity value, the curve complexity representation value is calculated as: 0.5 + 0.77 × 0.5 = 0.885. The weight here can be adjusted according to the actual scenario. Based on this high complexity representation value, the system automatically increases the sampling frequency of the vehicle camera (e.g., from 10 frames / second to 20 frames / second) and activates a high-precision steering decision algorithm to ensure that the vehicle can respond promptly to curve changes and pass smoothly when driving on S-shaped curves.

[0071] As shown above, this embodiment comprehensively calculates the curve quantification value using multi-dimensional indicators such as steering complexity, overall curvature, curvature variation, and curve curvature, and then combines this with a preset benchmark value to obtain an accurate curve complexity representation value. It also clarifies that the steering complexity representation value of the first type of curve is greater than that of the second type of curve, reflecting the complexity differences between different curve types. This enables the intelligent driving system to adjust task configuration according to the complexity characteristics of the curve, such as increasing the sensor sampling frequency and algorithm calculation accuracy in highly complex curves (e.g., multiple inflection points, large curvature, and dramatic curvature changes), ensuring driving safety.

[0072] Furthermore, considering that different roads (such as highways, urban roads, tunnels, etc.) have different requirements, such as speed limits and whether overtaking is permitted, this invention, based on the above embodiments, also combines different road conditions (such as straight roads, U-shaped curves, S-shaped curves, etc.) to set different weight coefficients for parameters such as steering complexity, overall curvature, curvature variation, and curve curvature. These weight coefficients can be fixed data or dynamically adjusted according to the actual scenario. This embodiment provides the implementation process for dynamically adjusting each weight coefficient:

[0073] Based on the driving scenario type, determine the steering frequency weight value, bending weight value, bending change weight value, and curve curvature weight value that match the driving scenario type; based on the steering complexity representation value and steering frequency weight value, overall bending degree representation value and bending weight value, bending change degree representation value and bending change weight value, and curve curvature value and curve curvature weight value corresponding to the first type of curve and the second type of curve, determine the curve quantification value.

[0074] The driving scenario type is a category categorized based on road attributes, including highways, closed urban roads, tunnels, and mountain roads. The turning frequency weight value W4 is a weight coefficient corresponding to the turning complexity representation value, used to adjust the influence of the turning frequency in the curve quantification value calculation. The curvature weight value W1 is a weight coefficient corresponding to the overall curvature representation value. The curvature variation weight value W2 is a weight coefficient corresponding to the curvature variation representation value. The curve curvature weight value W3 is a weight coefficient corresponding to the curve curvature value. For example, the image features closest to the current timestamp are obtained from the front-view image feature queue. This can be done using any network model with image classification capabilities; in this embodiment, it can be defined as a scene classification and detection network. This network first identifies the scenario type corresponding to the current track of the intelligent device, such as highways, tunnels, and closed urban roads. Then, the weight coefficients W1, W2, W3, and W4 are dynamically adjusted according to Table 1.

[0075] Table 1 Weighting Coefficients

[0076]

[0077] For example, at each point on the track fitting curve, the curvature corresponding to each pixel on the selected track line is calculated, the maximum curvature is counted, and then the curvature of each point is normalized. Then, the average curvature is obtained. (This parameter reflects the overall curvature of the driving lane), standard deviation of curvature (This parameter reflects the degree of abrupt change in curvature, i.e., whether there is a sharp turn), maximum curvature. (This parameter refers to the curvature of the sharpest curve in the lane, representing a high-risk moment.) Combined with preset straight-line complexity values ​​(such as 0.3 in the following formula), preset curve complexity values ​​(such as 0.7 in the following formula), and steering complexity values... Based on the respective weighting coefficients, the target trajectory complexity of the intelligent device in the curve scenario is obtained. for: .

[0078] As shown above, this embodiment dynamically adjusts the weighting coefficients according to the driving scenario type, making the calculation of curve quantification values ​​more aligned with the actual needs of different scenarios. For example, in highway scenarios, curve curvature has a higher weight, focusing on the risk of sharp curves; in urban road scenarios, the number of turns has a higher weight, adapting to the need for frequent turns. This improves the scenario adaptability of curve complexity assessment, making the task configuration of the intelligent driving system more targeted and further optimizing the driving experience.

[0079] Furthermore, this embodiment also provides an exemplary process for determining the complexity representation information of adjacent orbits, which may include the following:

[0080] Obtain the weighting coefficients corresponding to the track type; if the adjacent track line includes an exit ramp, and the driving action is not to leave the target track, the complexity representation information of the adjacent track is zero; if the driving action is to leave the target track, the complexity representation information of the adjacent track is determined based on the total number of tracks, the weighting coefficients corresponding to the track type, and the spatial position between the intelligent device and the exit ramp; if the adjacent track line includes an entry ramp, the complexity representation information of the adjacent track is determined based on the total number of tracks, the distance between the intelligent device and the entry ramp, the number of entry ramps, and the weighting coefficients corresponding to the track type.

[0081] The weighting coefficients are used to adjust the complexity calculation of adjacent tracks based on track type, and are divided into straight track weighting coefficients and curve weighting coefficients. An exit ramp refers to a ramp near the current travel track of the intelligent device that leads away from the entire travel track. An entry ramp is a ramp branching off from other roads used to enter the travel track of the intelligent device or an adjacent track. Spatial location refers to the relative distance and orientation between the intelligent device and the ramp. If there is an entry ramp to the left or right of the intelligent device, the farther the target track is from the entry ramp, the safer it is, and therefore the lower the complexity. If the intelligent device has an exit requirement, the closer the target track of the intelligent device is to the exit ramp, the fewer lanes it crosses, and the relatively lower the complexity. The total number of tracks refers to the total number of tracks (such as lanes) included in the current travel area.

[0082] For example, an autonomous vehicle is driving on a highway. There are currently four tracks, and the vehicle is on the second track (from left to right). The track type is a straight track, and the corresponding weighting coefficient is 0.5. Scenario 1: There is an exit ramp to the right of the vehicle, and the driving action is to leave the target track. Based on the total number of tracks (e.g., 4), the straight track weighting coefficient (e.g., 0.5), and the spatial distance between the vehicle and the exit ramp (e.g., 500 meters), the complexity representation of adjacent tracks is calculated as 0.5 × (2 / 4) × (1 - 500 / 1000) = 0.125 (the distance attenuation coefficient is set according to the actual scenario). Scenario 2: There is an entry ramp to the left of the vehicle, and the driving action is not to leave the target track. Based on the total number of tracks (e.g., 4), the distance between the vehicle and the entry ramp (e.g., 300 meters), the number of entry ramps (e.g., 1), and the straight-line weighting coefficient (e.g., 0.5), the adjacent track complexity representation is calculated as 0.5 × (2 / 4) × (300 / 1000) × 1 = 0.075. The system adjusts its configuration according to the adjacent track complexity representation in different scenarios: in scenario one, it increases the computational priority of lane-merging decisions and plans the lane-merging path in advance; in scenario two, it enhances the detection range of the left-side sensor to be vigilant against vehicles merging into the entry ramp, ensuring driving safety.

[0083] As can be seen from the above, this embodiment combines track type, driving action and ramp distribution (e.g., exit ramps and entry ramps) to comprehensively calculate the complexity representation information of adjacent tracks. When there is an exit ramp and the agent needs to leave, the influence of track position and spatial distance on lane merging is considered; when there is an entry ramp, the risks brought by the merging agent are taken into account. This achieves accurate assessment of the complexity of adjacent tracks and improves the safety and smoothness of lane merging and other operations.

[0084] For example, the present invention also provides a method for setting the weighting coefficients corresponding to track types, which may include the following: calculating the product of the curve length and the maximum curvature value of the track fitting curve to obtain the curvature product; calculating the sum of the curvature product and the preset adjustment parameter, using the ratio of the curvature product to the sum as the arc length weighted normalization weight, and determining the curve weighting coefficient based on the arc length weighted normalization weight and the preset weighting coefficient; if the track type is a straight track, then the weighting coefficient for the straight track type is the preset weighting coefficient; if the track type is not a straight track, then the weighting coefficient for the curve track type is the curve weighting coefficient.

[0085] Among them, the curve length refers to the actual length of the orbit fitting curve, which can be obtained by integral or piecewise calculation. The maximum curvature value is the maximum value among the curvatures of all points on the orbit fitting curve, reflecting the bending degree of the sharpest curve of the orbit. The curvature product is the product of the length of the orbit fitting curve and the maximum curvature value, used to reflect the comprehensive influence of the bending degree and length of the orbit. The preset adjustment parameter is a fixed parameter set to avoid abnormal weighting coefficients caused by too small curvature products. In this embodiment, it is set to be the same as the straight track weighting coefficient value. The arc length weighted normalization weight is a normalization weight calculated from the curvature product and the preset adjustment parameter, and its value range is between 0 and 1. The preset weighting coefficient is the set basic weighting coefficient. In this embodiment, it is set to 0.5. The curve track weighting coefficient is a weighting coefficient calculated for the curve track type and is used for the adjacent orbit complexity evaluation.

[0086] In this embodiment, considering that the curve track is more dangerous than the straight track, the weighting coefficient can be used for processing. The preset weighting coefficient is 0.5, and the preset adjustment parameter is 0.5. For the straight track, . For the curve track, , where p is the arc length weighted normalization weight, that is, for the orbit fitting curve, the middle of the position of the point with the maximum curvature, where L is the curve length, is the maximum curvature on the lane line. p has monotonicity (that is, the larger the curvature, the larger p), scale invariance (that is, when the curve is scaled, p remains unchanged), and normalization (p is between 0 and 1), so it can better describe the bending degree of different curves. For a straight line, p≈0; when the curve is slightly bent, 0<p<0.5; when the curve is significantly bent, 0.5<p<0.8; when the curve is severely bent, p≥0.8.

[0087] As can be seen from the above, in this embodiment, by calculating the product of the curve length and the maximum curvature value, combined with the preset adjustment parameter, the arc length weighted normalization weight is obtained, and then the curve track weighting coefficient is dynamically determined; the straight track adopts a fixed preset weighting coefficient, realizing the adaptation of the weighting coefficient to the actual characteristics of the orbit. It makes the calculation of the adjacent orbit complexity more in line with the real situation of the orbit, improves the accuracy of the complexity evaluation, and provides a more reliable basis for the intelligent driving task configuration.

[0088] For example, this embodiment also provides an exemplary calculation process for the complexity representation information of adjacent tracks: If the adjacent track line includes an exit ramp and the driving action is to leave the target track, then the exit ramp is taken as the starting position, the number of positions of the target track among all tracks is determined, the angle between the line segment between the location of the smart device and the center point of the exit ramp and the horizontal direction is calculated, and the complexity representation information of adjacent tracks is determined according to the ratio of the number of positions to the total number of tracks, the angle, and the track type weighting coefficient; If the adjacent track line includes an entry ramp, then the entry ramp is taken as the starting position, the number of positions of the target track among all tracks is determined, and the complexity representation information of adjacent tracks is determined according to the ratio of the number of positions to the total number of tracks, the preset complexity adjustment factor, the track type weighting coefficient, and the total number of entry ramps.

[0089] The center point of the exit ramp is the coordinate point at the point where the exit ramp separates from the main road. The horizontal direction is the horizontal reference direction parallel to the direction of travel of the intelligent device. The included angle is the angle between the line connecting the location of the intelligent device and the center point of the exit ramp and the horizontal direction, used to reflect the relative orientation of the vehicle and the ramp exit. The position number refers to the sequence number of the target track among all tracks in the current road (e.g., the 3rd track from left to right, position number 3). The total number of tracks is the total number of tracks included in the current driving space. The preset complexity adjustment factor is a fixed parameter used to adjust the degree of influence of the entering ramp on the complexity of adjacent tracks; in this embodiment, it is set to 0.3. The total number of entering ramps is the total number of entering ramps existing ahead of the current road.

[0090] In this embodiment, if there is an exit ramp to the left or right of the smart device, and the smart device does not require to exit, then the adjacent track complexity representation information value is taken. The value is 0.0; if the intelligent device has an exit requirement, the closer the intelligent device's lane is to the exit ramp, the fewer lanes it crosses, and the lower the complexity. Therefore, for intelligent devices with exit requirements, the calculation is performed according to the following steps. Taking the example of a lane with an exit ramp on the right side of the smart device lane, such as Figure 7 As shown, for exit ramps on the left, a similar approach can be used: obtain the current total number of lanes M (e.g., ...). Figure 8 (M is 4); for exiting the ramp on the right, the intelligent device lane is the Nth lane from right to left (e.g., N=3); calculate the angle between the location of the intelligent device and the center point of the ramp exit and the horizontal. , It can be represented as: .

[0091] If there is an on-ramp to the left or right in front of the intelligent device, the farther the intelligent device's lane is from the on-ramp, the safer it is, and therefore the lower the complexity. For example... Figure 8As shown, taking an on-ramp on the right as an example, the total number of lanes M (e.g.) Figure 8 (M is 4). For the case of entering the ramp from the right, if the intelligent device is in lane N from right to left, then in the case of entering the ramp from one side only... It can be represented as If there are entry lanes on both sides, then , Correspondingly, orbital complexity represents information. Possible forms: .

[0092] As shown above, this embodiment establishes refined complexity calculation models for both exit and entry ramps, introducing parameters such as the included angle, the ratio of the number of track positions to the total number, and preset adjustment factors. This achieves accurate quantification of the complexity of adjacent tracks in scenarios involving ramps. This enables the intelligent driving system to adjust task configurations based on the specific positional relationship between the vehicle and the ramp, such as optimizing lane-changing timing when approaching a ramp exit and providing early warnings of merging vehicles when there is an entry ramp, thus improving the safety and rationality of driving in ramp areas.

[0093] In addition to the complexity of the driving trajectory, the different environments in which intelligent devices operate place different requirements on intelligent driving systems. Based on the above embodiments, this embodiment also provides an implementation process for determining the driving environment type based on driving channel spatial data and determining scene complexity representation information based on the driving environment type, which may include the following:

[0094] From the spatial data of the driving channel, obtain the target image features within the time range of the target driving time; based on the target image features, determine the lighting type and weather type in front of the intelligent device to serve as the driving environment type; obtain the preset values ​​of lighting complexity and weather complexity respectively; based on the preset values ​​of lighting complexity corresponding to the lighting type and weather complexity corresponding to the weather type, determine the scene complexity representation information.

[0095] The target image features can be image features obtained from the front-view image feature queue; alternatively, only image features captured by the front and middle cameras can be acquired. Lighting type is a category categorized based on light intensity and angle, including daytime, nighttime, front lighting, and backlighting. Weather type is a category categorized based on weather conditions, including sunny, rainy, snowy, and foggy weather. The lighting complexity preset value is a pre-set quantification of complexity for different lighting types, used to reflect the degree of impact of lighting on driving. The weather complexity preset value is a pre-set quantification of complexity for different weather types, used to reflect the degree of impact of weather on driving.

[0096] For example, during the operation of an intelligent driving vehicle, target image features at the target driving time (such as the current time) are extracted from the spatial data of the driving lane. Based on the illumination classification network, the target image features are identified, determining the current illumination type to be nighttime backlight; based on the weather classification network, the current weather type is determined to be rainy. Preset values ​​for illumination complexity and weather complexity are retrieved: the preset illumination complexity value for nighttime backlight is 0.9, and the preset weather complexity value for rainy weather is 0.8. The scene complexity representation is calculated as: 0.9 × 0.8 = 0.72. The scene complexity representation uses a product method to comprehensively consider the effects of illumination and weather. Based on this scene complexity result, the system adjusts the task configuration: improving the computational accuracy of the image enhancement algorithm of the onboard camera, reducing the driving speed, and extending the braking distance to ensure driving safety in nighttime backlight and rainy weather scenarios.

[0097] As shown above, this embodiment accurately identifies lighting and weather types through target image features and, combined with preset complexity quantification values, achieves a quantitative assessment of scene complexity. This enables the intelligent driving system to adjust task configurations based on actual lighting and weather conditions, such as increasing the intensity of image enhancement algorithms in backlit scenarios and optimizing braking decision strategies in rainy scenarios, thereby improving the system's adaptability and safety under different environmental conditions.

[0098] Furthermore, this embodiment also provides an exemplary calculation process for scene complexity representation information, which may include obtaining preset values ​​for light intensity complexity corresponding to daytime and nighttime respectively; obtaining preset values ​​for illumination angle corresponding to front lighting and backlighting respectively; obtaining preset values ​​for weather complexity corresponding to different weather types; determining the illumination complexity value based on the preset values ​​for target illumination intensity and target illumination angle corresponding to the target image features; and using the product of the preset values ​​for target weather complexity and illumination complexity corresponding to the target image features as the scene complexity value.

[0099] The light intensity complexity preset value is a quantified value of complexity set based on light intensity (e.g., daytime, nighttime), used to distinguish the impact of different light intensities on driving. The illumination angle preset value is a quantified value of complexity set based on illumination angle (e.g., front lighting, backlighting), used to distinguish the impact of different illumination angles on driving. The target light intensity preset value is the light intensity complexity preset value matched to the light intensity type (i.e., daytime or nighttime) corresponding to the target image feature. The target illumination angle preset value is the illumination angle preset value matched to the illumination angle type (i.e., front lighting or backlighting) corresponding to the target image feature. The target weather complexity preset value is the weather complexity preset value matched to the weather type corresponding to the target image feature. The illumination complexity value is a quantified parameter of illumination complexity obtained by combining the target light intensity preset value and the target illumination angle preset value. The scene complexity value is a quantified parameter of the overall scene complexity obtained by combining the illumination complexity value and the target weather complexity preset value.

[0100] Obtain image features closest to the current timestamp and use a lighting classification network to identify daytime and nighttime conditions, as well as front lighting and backlighting. Preset value for lighting intensity complexity. For example, it could be: Preset value for illumination angle For example, it could be: Lighting complexity Using a weather classification network, rainy days, snowy days, sunny days, etc., are identified, with a preset weather complexity value. For example, it could be: Correspondingly, the scene complexity value It can be represented as: Ultimately, the complexity representation information of the driving trajectory scenario. It can be represented as: .

[0101] For example, preset lighting complexity parameters are: daytime light intensity complexity preset value is 0.5, nighttime is 1.0; front lighting angle preset value is 0.5, backlighting is 1.0. Preset weather complexity parameters are: sunny day 0.5, rainy day 0.8, snowy day 1.0. When the intelligent driving vehicle travels to a certain road segment, after extracting the target image features, it identifies the current lighting type as daytime backlighting and the weather type as snowy. The target light intensity preset value is determined to be 0.5 (i.e., the daytime light intensity complexity preset value), and the target lighting angle preset value is determined to be 1.0 (i.e., the backlighting angle preset value). The lighting complexity value is calculated using the average method: (0.5 + 1.0) / 2 = 0.75. The target weather complexity preset value is determined to be 1.0 (i.e., the preset weather complexity parameter corresponding to snowy days), and the scene complexity value is calculated: 0.75 × 1.0 = 0.75. The system adjusts the task configuration based on the complexity of the scenario: enabling snow driving mode, improving sensor anti-interference capabilities, optimizing anti-skid braking algorithm, reducing driving speed to a safe range, and ensuring driving safety in daytime backlight snow scenarios.

[0102] As shown above, this embodiment decomposes illumination complexity into two dimensions: illumination intensity and illumination angle, quantifying them separately, and then synthesizing them through a reasonable method to obtain the illumination complexity value. It uses a product method to fuse the illumination complexity value and the preset weather complexity value, accurately reflecting the synergistic impact of illumination and weather on scene complexity. This makes the scene complexity representation information more closely resemble the actual driving environment, providing a more accurate environmental assessment basis for the intelligent driving system, and further optimizing the targeting and rationality of task configuration.

[0103] It should be noted that there is no strict order of execution between the steps in this invention. As long as they conform to the logical order, these steps can be executed simultaneously or in a certain preset order. Figure 2This is just an illustrative example and does not mean that this is the only possible execution order.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0105] This invention also provides a corresponding apparatus for the intelligent driving task configuration method, further enhancing the practicality of the method. The apparatus can be described from both a functional module perspective and a hardware perspective. The intelligent driving task configuration apparatus provided by this invention is described below. This apparatus is used to implement the intelligent driving task configuration method provided by this invention. In this embodiment, the intelligent driving task configuration apparatus may include or be divided into one or more program modules. These one or more program modules are stored in a storage medium and executed by one or more processors to complete the intelligent driving task configuration method disclosed in Embodiment 1. The program module referred to in this embodiment refers to a series of computer program instruction segments capable of performing specific functions, which are more suitable than the program itself for describing the execution process of the intelligent driving task configuration apparatus in the storage medium. The following description will specifically introduce the functions of each program module in this embodiment. The intelligent driving task configuration apparatus described below can be referred to in correspondence with the intelligent driving task configuration method described above.

[0106] From the perspective of functional modules, the intelligent driving task configuration provided in this embodiment may include:

[0107] The data acquisition module is used to acquire spatial data of the driving lane ahead of the intelligent device.

[0108] The track complexity determination module is used to determine track line data based on the spatial data of the travel channel, and to determine the track complexity representation information of the intelligent device based on the track line data.

[0109] The scenario complexity determination module is used to determine the driving environment type based on the driving channel space data, and to determine the scenario complexity representation information based on the driving environment type.

[0110] The dynamic adaptation module is used to determine the driving trajectory scenario complexity representation information based on the trajectory complexity representation information and the scenario complexity representation information, and to adjust the intelligent driving task configuration information of the intelligent device based on the driving trajectory scenario complexity representation information.

[0111] For example, in some embodiments of this example, the above-mentioned track complexity determination module may be further used to: determine the target track line and adjacent track lines of the target track where the intelligent device is located based on track line data; perform curve fitting processing of no less than order three based on the position information of each track position point on the same side of the target track line to obtain a track fitting curve; determine the track type of the target track based on the extreme points and coefficients of the track fitting curve, and determine the target track complexity representation information based on the track type and the curvature information of the track fitting curve; determine the adjacent track complexity representation information in front of the intelligent device based on the track type of the target track, the driving action of the intelligent device and the adjacent track lines; and determine the track complexity representation information based on the target track complexity representation information and the adjacent track complexity representation information.

[0112] As an exemplary implementation of the above embodiments, the trajectory complexity determination module can also be used to: if the coefficients corresponding to the highest order and the second highest order of the trajectory fitting curve are both less than preset values, then the target trajectory is a straight track; if the target trajectory is not a straight track and the number of extreme points is not less than 2, then the target trajectory is a first type of curve with multiple inflection points; if the target trajectory is not a straight track and the number of extreme points is less than 2, then the target trajectory is a second type of curve with one inflection point.

[0113] As another exemplary implementation of the above embodiments, the track complexity determination module can also be used to: obtain preset straight track complexity values ​​and preset curve complexity values; determine the overall curvature representation value, curvature variation representation value, and curve curvature value of the target track based on the curvature information of the track fitting curve; determine the curve quantization value based on the steering complexity representation value, overall curvature representation value, curvature variation representation value, and curve curvature value corresponding to the first type of curve and the second type of curve; determine the curve complexity representation value based on the preset straight track complexity value, the preset curve complexity value, and the curve quantization value; if the target track is a straight track, the target track complexity representation information is the preset straight track complexity value; if the target track is not a straight track, the target track complexity representation information is the curve complexity representation value; wherein, the steering complexity representation value corresponding to the first type of curve is greater than the steering complexity representation value corresponding to the second type of curve.

[0114] As an exemplary implementation of the above embodiments, the track complexity determination module can also be used to: determine the turning frequency weight value, bending weight value, bending change weight value and curve curvature weight value that match the driving scenario type according to the driving scenario type; and determine the curve quantization value according to the turning complexity representation value and turning frequency weight value, overall bending degree representation value and bending weight value, bending change degree representation value and bending change weight value, and curve curvature value and curve curvature weight value corresponding to the first type of curve and the second type of curve.

[0115] As another exemplary implementation of the above embodiments, the track complexity determination module can also be used to: obtain the weighting coefficients corresponding to the track type; if the adjacent track line includes an exit ramp, and the driving action is not to leave the target track, then the adjacent track complexity representation information is zero; if the driving action is to leave the target track, then the adjacent track complexity representation information is determined according to the total number of tracks, the weighting coefficients corresponding to the track type, and the spatial position between the smart device and the exit ramp; if the adjacent track line includes an entry ramp, then the adjacent track complexity representation information is determined according to the total number of tracks, the distance between the smart device and the entry ramp, the number of entry ramps, and the weighting coefficients corresponding to the track type.

[0116] As another exemplary implementation of the above embodiments, the track complexity determination module can also be used to: calculate the product of the curve length and the maximum curvature value of the track fitting curve to obtain the curvature product; calculate the sum of the curvature product and the preset adjustment parameter, use the ratio of the curvature product to the sum as the arc length weighted normalization weight, and determine the curve weighting coefficient according to the arc length weighted normalization weight and the preset weighting coefficient; if the track type is a straight track, the weighting coefficient of the straight track type is the preset weighting coefficient; if the track type is not a straight track, the weighting coefficient of the curve track type is the curve weighting coefficient.

[0117] As another exemplary implementation of the above embodiments, the track complexity determination module can also be used to: if the adjacent track line includes an exit ramp and the driving action is to leave the target track, then take the exit ramp as the starting position, determine the number of positions of the target track among all tracks, calculate the angle between the line segment between the location of the smart device and the center point of the exit ramp and the horizontal direction, and determine the adjacent track complexity representation information based on the ratio of the number of positions to the total number of tracks, the angle, and the track type weighting coefficient; if the adjacent track line includes an entry ramp, then take the entry ramp as the starting position, determine the number of positions of the target track among all tracks, and determine the adjacent track complexity representation information based on the ratio of the number of positions to the total number of tracks, the preset complexity adjustment factor, the track type weighting coefficient, and the total number of entry ramps. For example, in some other embodiments of this embodiment, the data acquisition module described above may be further used to: acquire image data of the front of the smart device; acquire navigation data of the smart device; if the navigation data is missing or the driving track in the navigation data is missing, then the image data is used as driving channel space data; if the navigation data exists and the driving track information in the navigation data is complete, then the navigation data and the image data are used together as driving channel space data.

[0118] As an exemplary implementation of the above embodiments, the data acquisition module can also be used for: navigation data being navigation data of the smart device's navigation map in aerial view mode; processing the navigation data into layers according to different color layers to obtain sub-images of different color layers; performing binarization processing on each sub-image of each color layer; processing each binarized image using an edge algorithm to obtain track data; and determining track line topology information based on the track data; and determining the target track line topology information ahead of the smart device based on the track line topology information, according to the spatial location of the smart device, as track line data.

[0119] As another exemplary implementation of the above embodiments, the data acquisition module can also be used to: acquire images corresponding to multiple perspectives in front of the intelligent device while it is driving, and stitch the images together to form image data; extract image features from the image data and set acquisition time identifiers for the image features; and add each image feature to a pre-built front view image feature queue.

[0120] As another exemplary implementation of the above embodiments, the data acquisition module can also be used to: acquire target image features within the time range of the target's travel time from the front view image feature queue; identify two-dimensional track lines in the target image features; and convert the two-dimensional track lines into track lines under a bird's-eye view as track line data.

[0121] For example, in some other embodiments of this example, the scene complexity determination module may be further used to: obtain target image features within the time range of the target driving time from the driving channel space data; determine the lighting type and weather type in front of the intelligent device as the driving environment type based on the target image features; obtain preset values ​​for lighting complexity and weather complexity respectively; and determine scene complexity representation information based on the preset values ​​for lighting complexity corresponding to the lighting type and weather complexity corresponding to the weather type.

[0122] As an exemplary implementation of the above embodiments, the scene complexity determination module can also be used to: obtain preset values ​​of light intensity complexity corresponding to daytime and nighttime respectively; obtain preset values ​​of illumination angle corresponding to front lighting and backlighting respectively; obtain preset values ​​of weather complexity corresponding to different weather types; determine the illumination complexity value based on the preset values ​​of target illumination intensity and target illumination angle corresponding to the target image features; and use the product of the preset values ​​of target weather complexity and illumination complexity corresponding to the target image features as the scene complexity value.

[0123] For a description of the features in the embodiment corresponding to the intelligent driving task configuration device, please refer to the relevant description of the embodiment corresponding to the intelligent driving task configuration method, which will not be repeated here.

[0124] The intelligent driving task configuration device mentioned above is described from the perspective of functional modules. Furthermore, the present invention also provides an electronic device, which is described from a hardware perspective. This electronic device includes a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any of the above-described intelligent driving task configuration method embodiments.

[0125] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described embodiments of the intelligent driving task configuration method when it is run.

[0126] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0127] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described embodiments of the intelligent driving task configuration method.

[0128] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described embodiments of the intelligent driving task configuration method.

[0129] Finally, the present invention also provides an intelligent driving task configuration system, please refer to [link / reference]. Figure 9 The intelligent driving task configuration system includes at least an intelligent driving perception system 901 and multiple image acquisition devices 902 for collecting road data and environmental data ahead of the intelligent vehicle. The intelligent driving perception system 901 executes the steps of the intelligent driving task configuration method described in the above embodiments when executing a computer program stored in a memory.

[0130] The intelligent driving perception system refers to the system in an intelligent driving vehicle used to perceive the surrounding environment and road conditions. It includes sensor modules, data processing modules, and communication modules, responsible for collecting and processing spatial data of the driving path. Image acquisition equipment is hardware used to collect road and environmental image data, including forward-facing cameras and surround-view cameras, possessing multi-view acquisition and high-definition imaging capabilities. Road data refers to feature data related to the road, including track lines, curve shapes, and ramp locations. Environmental data refers to feature data related to the driving environment, including light intensity, weather conditions, and surrounding obstacles. By executing the above methods and steps, intelligent driving vehicles possess the ability to accurately assess scene complexity and dynamically adjust task configurations, improving the vehicle's adaptability to different road scenarios and driving safety.

[0131] Based on the intelligent driving task configuration system, when applied to fields such as roadside perception, intelligent transportation, V2X (Vehicle to Everything), intelligent security, and collaborative perception, such as... Figure 10 As shown, the process of dynamically adjusting intelligent driving task configuration information may include the following:

[0132] The intelligent device in this embodiment is an autonomous driving device. The autonomous vehicle (referred to as the autonomous vehicle in this embodiment) is equipped with an intelligent driving perception system consisting of LiDAR, cameras, and millimeter-wave radar, and also features six high-definition image acquisition devices (such as a front left camera, a front center camera, a front right camera, a left rear camera, a right rear camera, and a rear-view camera). During vehicle operation, the six image acquisition devices simultaneously acquire high-definition image data of the surrounding road and environment, while the LiDAR and millimeter-wave radar acquire three-dimensional road data and obstacle information. All data is transmitted to the intelligent driving perception system. The processor in the intelligent driving perception system executes a computer program to perform the following operations: fuse the collected data to obtain driving channel spatial data; extract track line data and driving environment type from the driving channel spatial data; calculate track complexity representation information, scene complexity representation information, and driving track scene complexity representation information; and adjust the task configuration based on the complexity results: in a sunny highway straight road scenario (corresponding to a low-complexity track scenario), reduce the sensor sampling frequency and algorithm calculation accuracy; in a rainy urban intersection curve scenario (corresponding to a high-complexity track scenario), increase the sensor sampling frequency, enable high-precision decision-making algorithms, and optimize braking and steering strategies. This intelligent driving vehicle, through the synergy of hardware configuration and software algorithms, achieves dynamic task configuration based on scene complexity, ensuring efficient and safe driving in different driving scenarios.

[0133] For example, when an autonomous vehicle is driving on a city expressway between 8:00 and 8:30 on a weekday, there is a right-side exit ramp 500 meters ahead, the weather is cloudy with good sunlight, lane markings are obscured in some sections due to construction, and the navigation map data is complete but needs to be combined with real-time images to correct local details, the process by which the processor in the intelligent driving perception system executes the computer program includes:

[0134] Step 1: Front view image stitching and feature queue construction.

[0135] Multi-view image acquisition: The vehicle is equipped with three high-definition cameras: front left (60° field of view), front center (90° field of view), and front right (60° field of view), which simultaneously acquire images of the road ahead at a frame rate of 15 frames per second. The SIFT feature matching algorithm is used to stitch the images from the three cameras into a single panoramic image with a field of view of 120°, solving the limitations of the single camera's field of view (such as the front center camera not being able to cover the rightmost exit ramp). The ResNet-18 backbone network is used to extract lane edge features, road surface texture features, and sky area illumination features from the stitched image. A collection timestamp (such as "2025-11-18 08:15:30") is added to each feature and added to the front view image feature queue. The queue capacity of the front view image feature queue is set to 50 entries, and the read / write mode is first-in-first-out.

[0136] Step 2: Obtain lane topology information from the BEV perspective.

[0137] Navigation map preprocessing: The navigation map is invoked and switched to "Aerial View Mode (BEV) + Vehicle Facing Up" to avoid unnecessary calculations during subsequent perspective transitions. In the navigation map, lane lines are white (RGB values: 255, 255, 255), road surfaces are light gray (RGB values: 240, 240, 240), and median strips are green (RGB values: 0, 200, 0). Three sub-images are obtained based on color layering. The white sub-image (i.e., the tile corresponding to the lane lines) is binarized: pixel values ​​≥ 250 are set to 25. 5, otherwise 0; The Sobel edge algorithm is used to extract the lane line edges of the binarized image to obtain the topology of "4 main lanes + 1 right exit ramp". Combined with the current position of the vehicle, the lane line topology within a range of 500 meters ahead is extracted, and the lane where the vehicle is currently located is marked as "the second main lane from the right". The image features at the current moment are retrieved from the front view image feature queue. It is found that 20 meters of lane lines at the entrance of the exit ramp are blocked due to construction. The lane lines in the blocked area are filled in by the local features of the front right camera to ensure the integrity of the BEV topology.

[0138] Step 3: Lane complexity calculation.

[0139] One hundred location points were extracted from the right lane line of the vehicle lane, with each location point spaced 5 meters apart. These points were then fitted into a cubic curve equation, yielding coefficients a=0.0002, b=0.001, c=0.1, and d=5. Since a and b are not close to 0, the curve is identified as a curve. Differentiating the fitted curve, the number of extreme points is 1, indicating a U-shaped curve. Correspondingly, the turning complexity Cturns=0.5. The curvature at each point on the fitted curve is calculated, yielding an average curvature Cavg_curv = 0.6, a standard deviation of curvature Csd_curv = 0.2, and a maximum curvature Cmax_curv = 0.8. The scene classification network identifies the current location as an "urban expressway," and weights w1 = 0.2, w2 = 0.3, w3 = 0.3, and w4 = 0.2 are set. The target track complexity value is determined using Cego_road = 0.3 + 0.7 × 40.2 × 0.6 + 0.3 × 0.2 + 0.3 × 0.8 + 0.2 × 0.5 = 0.3 + 0.7 × 0.475 = 0.6325.

[0140] The vehicle lane is a curve, with a curve length L = 500 meters and a maximum curvature kmax = 0.8. The curve is weighted and normalized using arc length. Therefore, the curve weighting coefficient wother_road = 0.5 + 0.5 × 0.97 ≈ 0.985; the vehicle needs to leave the target track and enter the right ramp. The total number of lanes M = 4, and the number of track positions of the vehicle's lane from right to left N = 2. The angle θ = 25° between the line connecting the vehicle and the center point of the ramp exit and the horizontal is calculated. The complexity value of the adjacent track is obtained by Cother_road = 0.985 × 42 × sin25° ≈ 0.985 × 0.5 × 0.4226 ≈ 0.208. The overall lane complexity Croad = 20.6325 + 0.208 ≈ 0.420.

[0141] Step 4: Calculate the environmental complexity.

[0142] The sky region features of the current image are retrieved from the feature queue. The illumination classification network determines it as "daytime with front lighting," with Cday_or_night = 0.5 and Cbacklight_or_frontlight = 0.5. Therefore, Cillumination = (0.5 + 0.5) / 2 = 0.5. Based on the road surface wet texture and sky grayscale features in the image, the weather classification network determines it as "cloudy with no rain," with Cweather, for example, being 0.6. The overall environmental complexity, Cenvironment = 0.5 × 0.6 = 0.3.

[0143] Step 5: Output the total static complexity of the road.

[0144] The total complexity Call = Croad × Center = 0.420 × 0.3 ≈ 0.126, which corresponds to a low complexity level.

[0145] Based on a total complexity of 0.126 (i.e., a low-complexity track scenario), the intelligent driving system performs the following configuration adjustments: Sensors: Reduce the LiDAR sampling frequency (from 10Hz to 5Hz), while retaining the basic frame rate of 15 frames / second for the forward-looking camera; Algorithm: Replace the high-precision but time-consuming Lane-ATT with the lightweight model ultra-Fast-lane-De for lane line recognition; Decision: Since a smooth lane change is required in the ramp scenario, maintain the current speed of 60km / h, activate the right turn signal 300 meters in advance, and there is no need to activate the emergency braking plan.

[0146] As shown above, this embodiment proposes a fast method for evaluating the static complexity of road scenes. In terms of lane complexity analysis, if a local navigation map is available, lane lines from the BEV perspective are obtained through color layering and edge operators. For cases where a navigation map is unavailable, images from the front-view cameras (e.g., front left, front center, front right) are stitched together to obtain a larger image area. Image features are then extracted from the stitched image to obtain 2D lane lines in the image coordinate system, and IPM transformation is used to obtain the lane lines from the BEV perspective. For lane lines from the BEV perspective, the complexity of the lane where the vehicle is located is first analyzed (e.g., straight or curved), and then the complexity of adjacent lanes is analyzed. Regarding environmental complexity evaluation, the complexity of lighting environment, weather, and scene are identified and evaluated based on the features of the stitched image at the current moment. Finally, combined with lane complexity, a static road complexity evaluation is given, providing support for subsequent algorithm / model selection.

[0147] The intelligent driving task configuration method and electronic device provided by this invention have been described in detail above. The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Whether the units and algorithm steps of each example described in the disclosed embodiments are executed by electronic hardware or computer software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, and such implementations should not be considered beyond the scope of this invention. Several improvements and modifications can be made to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of this invention.

Claims

1. A method for intelligent driving task configuration, characterized in that, The method comprises the following steps: acquiring driving channel space data in front of the intelligent device on a driving route; determining track line data according to the driving channel space data, and determining track complexity representation information of the intelligent device according to the track line data; determining driving environment types according to the driving channel space data, and determining scene complexity representation information according to the driving environment types; determining driving track scene complexity representation information according to the track complexity representation information and the scene complexity representation information, and adjusting intelligent driving task configuration information of the intelligent device according to the driving track scene complexity representation information; wherein the track complexity representation information at least comprises target track complexity representation information, and the determination process of the target track complexity representation information comprises: determining a target track line and adjacent track lines of a target track where the intelligent device is located according to the track line data; performing curve fitting processing of not less than three orders according to position information of each track position point where the target track line is located on the same side to obtain a track fitting curve; and determining a track type of the target track according to extreme points and coefficients of the track fitting curve; acquiring a preset straight track complexity value and a preset curved track complexity value; determining an overall bending degree representation value, a bending change degree representation value and a curved track curvature value of the target track according to curvature information of the track fitting curve; determining a curved track quantization value according to a steering complexity representation value corresponding to a first type curved track and a second type curved track, the overall bending degree representation value, the bending change degree representation value and the curved track curvature value; and determining a curved track complexity representation value according to the preset straight track complexity value, the preset curved track complexity value and the curved track quantization value; the steering complexity representation value corresponding to the first type curved track is greater than the steering complexity representation value corresponding to the second type curved track; if the target track is a straight track, the target track complexity representation information is the preset straight track complexity value; if the target track is not a straight track, the target track complexity representation information is the curved track complexity representation value.

2. The intelligent driving task configuration method of claim 1, wherein, determining track complexity representation information of the intelligent device according to the track line data comprises: determining target track complexity representation information according to the track type and the curvature information of the track fitting curve; determining adjacent track complexity representation information in front of the intelligent device according to the track type of the target track, driving actions of the intelligent device and adjacent track lines; determining the track complexity representation information according to the target track complexity representation information and the adjacent track complexity representation information.

3. The intelligent driving task configuration method of claim 2, wherein, determining the track type of the target track according to the extreme points and the coefficients of the track fitting curve comprises: if coefficients corresponding to the highest order and the second highest order of the track fitting curve are both less than a preset value, the target track is a straight track; if the target track is not a straight track and the number of extreme points is not less than 2, the target track is a first type curved track with multiple inflection points; if the target track is not a straight track and the number of extreme points is less than 2, the target track is a second type curved track with one inflection point. 4.The intelligent driving task configuration method of claim 1, wherein, The turning complexity representation value corresponding to the first type of curve and the second type of curve, the overall bending degree representation value, the bending change degree representation value and the curve curvature value are used to determine a curve quantization value, including: According to the driving scene type, determine the turning frequency weight value, the bending weight value, the bending change weight value and the curve curvature weight value matched with the driving scene type; According to the turning complexity representation value corresponding to the first type of curve and the second type of curve and the turning frequency weight value, the overall bending degree representation value and the bending weight value, the bending change degree representation value and the bending change weight value, and the curve curvature value and the curve curvature weight value, determine the curve quantization value.

5. The intelligent driving task configuration method of claim 2, wherein, According to the track type of the target track, the driving action of the intelligent device and the adjacent track line, determine the adjacent track complexity representation information in front of the intelligent device during driving, including: Obtain the weighting coefficient corresponding to the track type of the target track; If the adjacent track line includes an exit ramp, if the driving action is not to drive off the target track, the adjacent track complexity representation information is zero; if the driving action is to drive off the target track, determine the adjacent track complexity representation information according to the total number of tracks, the weighting coefficient corresponding to the track type, and the spatial position between the intelligent device and the exit ramp; If the adjacent track line includes an entrance ramp, determine the adjacent track complexity representation information according to the total number of tracks, the distance between the intelligent device and the entrance ramp, the number of entrance ramps and the weighting coefficient corresponding to the track type; The weighting coefficient corresponding to the track type is: Calculate the product of the curve length and the maximum curvature value of the track fitting curve to obtain a curvature product; calculate the sum of the curvature product and a preset adjustment parameter, take the ratio of the curvature product and the sum as an arc length weighted normalization weight, and determine a curve weighting coefficient according to the arc length weighted normalization weight and a preset weighting coefficient; if the track type is a straight track, the weighting coefficient of the straight track type is the preset weighting coefficient; if the track type is not a straight track, the weighting coefficient of the curve track type is the curve weighting coefficient. 6.The intelligent driving task configuration method of claim 2, wherein, According to the track type of the target track, the driving action of the intelligent device and the adjacent track line, determine the adjacent track complexity representation information in front of the intelligent device during driving, including: If the adjacent track line includes an exit ramp and the driving action is to drive off the target track, take the exit ramp as the starting position, determine the number of positions of the target track in all tracks, calculate the angle between the line segment between the position of the intelligent device and the center point of the exit ramp outlet and the horizontal direction, and determine the adjacent track complexity representation information according to the ratio of the number of positions to the total number of tracks, the angle and the weighting coefficient corresponding to the track type. If the adjacent track line includes a driving-in ramp, the driving-in ramp is taken as a starting position, the number of positions where the target track is located in all tracks is determined, and the adjacent track complexity representation information is determined according to a ratio of the number of positions to a total number of tracks, a preset complexity adjustment factor, a weighting coefficient corresponding to a track type, and a total number of driving-in ramps. 7.The intelligent driving task configuration method of claim 1, wherein, Obtaining driving channel space data in front of the intelligent device on a driving route, including: Obtaining image data in front of the intelligent device; Obtaining navigation data of the intelligent device; If the navigation data does not exist or the driving track in the navigation data is missing, taking the image data as the driving channel space data; If the navigation data exists and the driving track information in the navigation data is complete, taking the navigation data and the image data together as the driving channel space data. 8.The intelligent driving task configuration method of any one of claims 1 to 7, characterized in that, Determining a driving environment type according to the driving channel space data, and determining scene complexity representation information according to the driving environment type, including: Obtaining target image features within a time range of a target driving time from the driving channel space data; Determining a light type and a weather type in front of the intelligent device as the driving environment type according to the target image features; Respectively obtaining a preset light complexity value and a preset weather complexity value; Determining the scene complexity representation information according to the preset light complexity value corresponding to the light type and the preset weather complexity value corresponding to the weather type.

9. An electronic device, comprising: Including: A memory for storing a computer program; A processor for executing the computer program to realize the steps of the intelligent driving task configuration method according to any one of claims 1 to 8.

Images