Vehicle path planning method and device based on traffic flow and storage medium

Abstract

The invention discloses a traffic flow-based vehicle path planning method and device, and a storage medium. The method comprises the steps of obtaining a traffic flow from a plurality of vehicle trajectories generated by a vehicle-mounted tracking module; the traffic flow is preprocessed; converting the preprocessed traffic flow into a traffic flow field with a multi-channel structure; and performing path planning based on the traffic flow field. According to the method, human-like path planning can be realized by using the traffic flow field without using a high-definition map, so that the marking cost is greatly reduced. Potential map errors can be detected by using the traffic flow field, and a high-definition map can be timely fed back and updated.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle path planning, and in particular relates to a vehicle path planning method, device and storage medium based on traffic flow. Background Art

[0002] Autonomous driving currently relies mainly on single-vehicle autonomous driving (AD), which is mainly divided into several modules, such as positioning, perception, prediction, decision planning and control execution.

[0003] Currently, one of the autonomous driving planning schemes for highways or urban driving relies on high-precision maps, that is, the map needs to provide lane-level navigation information for lane-level planning.

[0004] The high-precision map-based autonomous driving planning solution requires the following modules to generate high-precision maps for planning:

[0005] 1) Data collection: Use various sensors and measuring equipment to collect detailed data of the road, including road shape, lane lines, traffic signs, signal lights, road signs, etc.

[0006] 2) Data processing and integration: Process the collected data, integrate it into a unified format, and ensure the accuracy and completeness of the data.

[0007] 3) Map production: Use the processed data to create high-precision maps, including static maps and dynamic maps. Static maps contain fixed features of the road, while dynamic maps contain traffic information that changes in real time.

[0008] 4) Accuracy calibration: Ensure the accuracy of map data, which usually requires centimeter-level accuracy to meet the high requirements of autonomous vehicles for position accuracy.

[0009] 5) Map Updates: Map data is updated regularly to reflect changes in road and traffic conditions, such as construction, traffic sign updates, etc.

[0010] Autonomous driving planning solutions that rely on high-precision maps have the following disadvantages:

[0011] 1) High updating and maintenance costs: High-precision maps need to be updated regularly to reflect road changes, which involves high updating and maintenance costs.

[0012] 2) Limited coverage: High-precision maps may only have detailed data in certain specific areas or cities, resulting in limited autonomous driving functions in uncovered areas.

[0013] 3) Difficulties in data collection and processing: The surveying and mapping of high-precision maps requires a large amount of data collection work and extremely high precision requirements, which increases the difficulties in data collection and processing.

[0014] 4) Dependence on sensors: Although high-precision maps provide auxiliary information, the autonomous driving system still needs to rely on in-vehicle sensors to perceive the surrounding environment, which may lead to sensor failures or limitations in some cases.

[0015] 5) Legal and policy restrictions: High-precision maps involve spatial information security and may be restricted by laws and policies, affecting their development and application.

[0016] 6) Adaptability to dynamic environments: High-precision maps may have difficulty reflecting temporary changes in roads in a timely manner, such as construction and traffic control, which requires the autonomous driving system to be able to quickly adapt to dynamic environments.

[0017] Another lane-level planning scheme is autonomous driving without a high-definition map relying on online perception, that is, the technology of emphasizing perception over maps.

[0018] Benefiting from the latest advancements in neural network architectures such as Transformer and Bird's Eye View (BEV) representation, the image features from surrounding cameras and / or the

[0019] point clouds from LiDAR are encoded, and the lane prediction problem is modeled in the form of a Bird's Eye View (BEV) representation. Multi-head attention is used to correctly find the corresponding features from the perspective view or the LiDAR point cloud, and the vectorized map elements in the BEV are predicted. Additionally, a Transformer-based semantic learning method is used to simultaneously detect objects and segment the road structure to detect the road topology information in real time for downstream autonomous driving planning.

[0020] The autonomous driving planning scheme relying on online perception without a high-definition map has the following disadvantages:

[0021] 1) Perception range limitation: Vehicle sensors, such as cameras, radars, and LiDAR, usually have a limited perception range and field of view, and may not be able to perceive the environment over long distances or omnidirectionally.

[0022] 2) Influence of environmental factors: The performance of sensors may be affected by adverse weather conditions (such as rain, snow, fog) or lighting changes (such as at night or against the light).

[0023] 3) Data processing and computing power requirements: Online perception requires real-time processing of a large amount of sensor data, which requires the vehicle to have powerful computing capabilities to react quickly.

[0024] 3) Data processing and computing power requirements: Online perception requires real-time processing of a large amount of sensor data, which requires the vehicle to have powerful computing capabilities to react quickly.

[0025] 4) Sensor fusion challenges: Different types of sensors may provide different information, and complex data fusion algorithms are required to integrate this information to obtain an accurate understanding of the environment.

[0026] 5) Corner Case handling: For rare or extreme traffic scenarios (Corner Cases), online perception may have difficulty in timely identifying and making correct responses.

[0027] 6) Lack of dynamic traffic information: The light map solution may lack real-time updates of dynamic traffic information, such as traffic accidents, construction, or temporary traffic control.

[0028] 7) Dependence on real-time data streams: Online perception solutions highly depend on real-time data streams, and any sensor failure or data transmission interruption may affect the stability of the autonomous driving system.

[0029] 8) Safety and redundancy issues: Online perception solutions may lack sufficient safety redundancy mechanisms. Once the main sensor system fails, there may be no backup system to take over.

[0030] 9) Cost and complexity: Although the dependence on high-precision maps is reduced, the vehicle needs to be equipped with more high-performance sensors and computing devices, which may increase costs and system complexity. Summary of the Invention

[0031] To solve the above technical problems, the present invention proposes a vehicle path planning method, device, and storage medium based on traffic flow.

[0032] To achieve the above object, the technical solution of the present invention is as follows:

[0033] In a first aspect, the present invention discloses a vehicle path planning method based on traffic flow, including:

[0034] Step S1: Obtain traffic flow from a number of vehicle trajectories generated by an in-vehicle tracking module;

[0035] Step S2: Preprocess the traffic flow;

[0036] Step S3: Convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure;

[0037] Step S4: Perform path planning based on the traffic flow field.

[0038] Based on the above technical solution, the following improvements can be made:

[0039] As a preferred solution, step S2 includes:

[0040] Step S2.1: Filter out the unqualified vehicle trajectories in the traffic flow based on one or more factors such as the tracked life cycle, the total travel distance of the trajectory, whether there is abnormal drift, and whether there is a collision with an obstacle;

[0041] Step S2.2: Further filter the traffic flow to screen out the vehicle trajectories with clear entry points and exit points relative to the region of interest. Each vehicle trajectory carries the edge information corresponding to the region of interest, and obtain

[0042] F = b v , m v |v ∈ V;

[0043] m v =(l entry , l exit , p start );

[0044] Where: F is the set of vehicle trajectories;

[0045] b v is the path of vehicle v at all tracked timestamps;

[0046] V is the set of vehicle tracking ids;

[0047] m v is the information carried by the vehicle trajectory;

[0048] l entry and l exit are the entry edge and exit edge corresponding to the region of interest respectively;

[0049] p start is the intersection point of l entry .

[0050] As a preferred solution, step S3 includes:

[0051] Step S3.1: Group the vehicle trajectories in the traffic flow according to the edge information corresponding to the region of interest carried by each vehicle trajectory. The vehicle trajectories within the group have the same entry and exit points;

[0052] Step S3.2: Obtain multiple entry points p e , p e for each group of vehicle trajectories through clustering. p is the channel-level identifier for each vehicle trajectory;

[0053] Step S3.3: Group according to the entry point p e to obtain the flow rates corresponding to different entry points

[0054]

[0055] Step S3.4: Construct a traffic flow field with a multi-channel structure, where each channel represents the traffic flow from a specific entry point.

[0056] Step S3.5: For each channel of the traffic flow field, construct a grid, and each cell of the grid stores the information f i,j =(d, δ i , δ j );;

[0057] where: d represents the traffic density, that is, the number of vehicles passing through this cell;

[0058] δ i , δ j are the average directions at cells i and j respectively.

[0059] As a preferred solution, Step S3.1 includes:

[0060] First, extract the edge information m corresponding to the region of interest carried by each trajectory v ;

[0061] Then, based on the entry edge l v and the exit edge l entry in m exit , generate the entry target g in and the exit target g out ;

[0062] Finally, group the set F of vehicle trajectories based on the entry target g in and the exit target g out . The vehicle trajectories within the group have the same entry target g in and the same exit target g out .

[0063] As a preferred solution, Step S3.5 further includes the following: Each region of interest in the traffic flow field is designed as a separate grid.

[0064] As a preferred solution, under the continuous update of traffic flow information, each region of interest in the traffic flow field is updated based on the first-in-first-out queue mechanism.

[0065] As a preferred solution, Step S4 includes:

[0066] Step S4.1: Load the traffic flow field;

[0067] Step S4.2: Enter the region of interest to query the traffic flow field data;

[0068] Step S4.3: Search for the optimal initial path based on the traffic density and the average direction of the traffic flow;

[0069] Step S4.4: Establish a frenet coordinate system based on the optimal initial path;

[0070] Step S4.5: Sample the longitudinal nodes and the corresponding lateral clusters based on the frenet coordinate system;

[0071] Step S4.6: Based on the graph formed by the longitudinal nodes and the lateral clusters, use the dynamic programming algorithm to generate multiple potential candidate paths;

[0072] Step S4.7: Remove the duplicate paths from the candidate paths;

[0073] Step S4.8: Perform quadratic programming optimization on the candidate paths after removing the duplicate paths to smooth the paths.

[0074] In a second aspect, the present invention discloses a vehicle path planning device based on traffic flow, including:

[0075] An acquisition module, configured to acquire traffic flow from a plurality of vehicle trajectories generated by an on-vehicle tracking module;

[0076] A preprocessing module, configured to preprocess the traffic flow;

[0077] A conversion module, configured to convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure;

[0078] A path planning module, configured to perform path planning based on the traffic flow field.

[0079] As a preferred solution, the preprocessing module includes:

[0080] A preliminary filtering unit, configured to filter out unqualified vehicle trajectories in the traffic flow based on one or more factors such as the tracking life cycle, the total travel distance of the trajectory, whether there is abnormal drift, and whether there is a collision with an obstacle;

[0081] A secondary filtering unit, configured to further filter the traffic flow and screen out vehicle trajectories with clear entry points and exit points relative to the region of interest, and each vehicle trajectory carries edge information corresponding to the region of interest, to obtain

[0082] F = b v , m v |v ∈ V;

[0083] m v = (l entry , l exit , p start );

[0084] Wherein: F is the set of vehicle trajectories;

[0085] b vThe path of vehicle v at all timestamps being tracked;

[0086] V is the set of vehicle tracking IDs;

[0087] m v Is the information carried by the vehicle trajectory;

[0088] l entry And l exit Are respectively the entrance edge and the exit edge corresponding to the region of interest;

[0089] p start Is the intersection point of l entry ;

[0090] As a preferred solution, the conversion module includes:

[0091] The first grouping unit is used to group the vehicle trajectories of the traffic flow according to the edge information corresponding to the region of interest carried by each vehicle trajectory, and the vehicle trajectories within the group have the same entrance and exit;

[0092] The clustering unit is used to obtain multiple entry points p through clustering for each group of vehicle trajectories e , p e Is the channel-level identifier of each vehicle trajectory;

[0093] The second grouping unit is used to group according to the entry point p e To obtain the traffic flow corresponding to different entry points

[0094] The traffic flow field construction unit is used to construct a traffic flow field with a multi-channel structure, and each channel is the traffic flow from a specific entry point

[0095] The grid construction unit is used to construct a grid for each channel of the traffic flow field, and each cell of the grid stores the information f i,j =(d, δ i , δ j );

[0096] Where: d represents the traffic flow density, that is, the number of vehicles passing through this cell;

[0097] δ i , δ j Are respectively the average directions at cells i, j.

[0098] As a preferred solution, the first grouping unit is used to execute the following method, including:

[0099] First, extract the edge information m corresponding to the region of interest carried by each trajectory v ;

[0100] Then, according to m v the inlet edge l entry and the outlet edge l exit generate the inlet target g in and the outlet target g out ;

[0101] Finally, based on the inlet target g in and the outlet target g out group the set F of vehicle trajectories, and the vehicle trajectories within the group have the same inlet target g in and the outlet target g out .

[0102] As a preferred solution, the grid construction unit is further configured to perform the following: Each region of interest in the traffic flow field is designed as a separate grid.

[0103] As a preferred solution, under the continuous update of traffic flow information, each region of interest in the traffic flow field is updated based on the first-in-first-out queue mechanism.

[0104] As a preferred solution, the path planning module includes:

[0105] A loading unit for loading the traffic flow field;

[0106] A query unit for querying traffic flow field data by entering the region of interest;

[0107] An initial path search unit for searching for the optimal initial path based on the traffic density and the average direction of the traffic flow;

[0108] A coordinate system establishment unit for establishing a frenet coordinate system according to the optimal initial path;

[0109] A sampling unit for sampling the longitudinal nodes and the corresponding transverse clusters of the longitudinal nodes based on the frenet coordinate system;

[0110] A dynamic programming unit for generating multiple potential candidate paths by using the dynamic programming algorithm based on the graph formed by the longitudinal nodes and the transverse clusters;

[0111] A duplicate path removal unit for removing duplicate paths in the candidate paths;

[0112] A path smoothing unit for performing quadratic programming optimization on the candidate paths after removing duplicate paths to smooth the paths.

[0113] In addition, in a third aspect, the present invention also discloses a storage medium storing one or more computer-readable programs, and the one or more programs include instructions adapted to be loaded and executed by a memory to perform any of the above vehicle path planning methods.

[0114] A vehicle path planning method, device, and storage medium according to the present invention have the following beneficial effects:

[0115] First, for extra-large and irregular intersections, using a traffic flow field can achieve human-like path planning without using a high-definition map, greatly reducing the annotation cost.

[0116] Second, since its incremental update adopts a first-in, first-out vehicle flow update strategy, it has good adaptability to change scenarios such as road construction and road routing changes, enabling the intelligent driving algorithm to update the planning result in a timely manner according to environmental changes.

[0117] Third, a traffic flow field can be used to detect potential map errors and timely feedback to update the high-definition map. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0119] Figure 1 It is a flowchart of the vehicle path planning method provided by the embodiment of the present invention.

[0120] Figure 2 It is a schematic diagram of vehicle trajectories at an intersection provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0121] The following will describe in detail the preferred embodiments of the present invention with reference to the drawings.

[0122] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0123] Using ordinal numbers such as "first", "second", "third", etc. to describe ordinary objects only represents different instances of similar objects, and does not intend to imply that the objects so described must have a given order in terms of time, space, sorting, or any other way.

[0124] In addition, the expression "comprising" an element is an "open-ended" expression, which merely means that the corresponding components or steps exist and should not be construed as excluding additional components or steps.

[0125] Human drivers do not drive solely based on lane or road markings. When human drivers come to unknown places, they tend to follow the tracks of other human drivers. For example, at intersections during urban commuting, when passing through the intersection multiple times, the traffic flow at the intersection can be accumulated, and the traffic flow consists of many vehicle trajectories generated by the on-vehicle tracking module.

[0126] The traffic flow contains rich driving behaviors and a large amount of information about multi-modal human driving behaviors, but it is unstructured. To facilitate planning and control, the present invention proposes a vehicle path planning method to extract human-like driving paths from the traffic flow field. Based on this method, guiding paths can be generated even in large open spaces with few road markings.

[0127] To achieve the object of the present invention, in some embodiments of a vehicle path planning method, device, and storage medium based on traffic flow, as Figure 1 shown, the vehicle path planning method includes:

[0128] Step S1: Obtain traffic flow from several vehicle trajectories generated by the on-vehicle tracking module;

[0129] Step S2: Preprocess the traffic flow;

[0130] Step S3: Convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure;

[0131] Step S4: Perform path planning based on the traffic flow field.

[0132] Each step will be elaborated in detail below.

[0133] In step S1, traffic flow is obtained from several vehicle trajectories generated by the on-vehicle tracking module.

[0134] The input of the traffic flow comes from the on-vehicle tracking module, represented as the tracking result of a specific tracked vehicle with ID i at timestamp t; p = xyθ represents the center coordinates and orientation of the tracked bounding box, and l and w respectively represent the length and width of the bounding box.

[0135] The set of all observed vehicle trajectories is represented as follows:

[0136] F = b v | v ∈ V;

[0137] Where: b v is the path of vehicle v at all timestamps during tracking;

[0138] V is a set of vehicle tracking IDs.

[0139] Note that these vehicle trajectories can be obtained by a single vehicle or a group of vehicles at different spaces and times. Figure 2 Shows an example of F under a specific intersection.

[0140] The generation of traffic flow is lightweight and only depends on on-vehicle tracking.

[0141] However, there are the following three main problems:

[0142] (1) Due to occlusion, the lifespan of many observed trajectories is very short.

[0143] (2) There may be perception errors that can cause abnormal trajectories, such as trajectory drift or collision with obstacles.

[0144] (3) There are stationary or short trajectories that contain limited information.

[0145] Therefore, it is necessary to filter F so that it only contains high-quality vehicle trajectories.

[0146] Step S2, preprocess the traffic flow.

[0147] Furthermore, step S2 includes:

[0148] Step S2.1: Based on four factors, namely the life cycle of tracking, the total travel distance of the trajectory, whether there is abnormal drift, and whether there is a collision with an obstacle, filter out unqualified vehicle trajectories in the traffic flow;

[0149] Step S2.2: Further filter the traffic flow to screen out vehicle trajectories with clear entry and exit points relative to the region of interest (ROI). Each vehicle trajectory carries the edge information corresponding to the region of interest, and obtain

[0150] F = b v , m v | v ∈ V;

[0151] m v = (l entry , l exit , p start );

[0152] Where: F is a set of vehicle trajectories;

[0153] b v is the path of vehicle v at all timestamps of tracking;

[0154] V is a set of vehicle tracking IDs;

[0155] m vCarry information for vehicle trajectories;

[0156] l entry and l exit are respectively the entry edge and the exit edge corresponding to the region of interest (i.e., the edge that intersects with the traffic flow trajectory at the intersection);

[0157] p start is the intersection point of l entry .

[0158] The traffic flow F is essentially a set of vehicle trajectories organized by spatial position. Since no explicit structure is constructed, F cannot be directly used for path generation. Therefore, it is necessary to generate a traffic flow field G based on the traffic flow F, which encapsulates the flow density and flow direction information in a multi-channel grid map. However, the flows from different directions will be chaotic with each other. And the flows from different directions are often unbalanced.

[0159] Step S3: Convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure.

[0160] The reason for extracting lane-level information is that there are inherent multimodals in the traffic flow, even with the same entry and exit targets.

[0161] Furthermore, step S3 includes:

[0162] Step S3.1: Group the vehicle trajectories of the traffic flow according to the edge information corresponding to the region of interest carried by each vehicle trajectory. The vehicle trajectories within a group have the same entry and exit;

[0163] Step S3.2: Obtain multiple entry points p e , p e for each group of vehicle trajectories through clustering.

[0164] Step S3.3: Group according to the entry point p e to obtain the flow corresponding to different entry points

[0165]

[0166] Step S3.4: Construct a traffic flow field with a multi-channel structure, where each channel is the flow from a specific entry point ( which can be understood as a subset of the traffic flow constructed with the lane of the entry point as the index);

[0167] Step S3.5: For each channel of the traffic flow field, construct a grid, and each cell of the grid stores the information f i,j =(d, δ i , δ j );;

[0168] Where: d represents the traffic density, i.e., the number of vehicles passing through the cell;

[0169] δ i and δ j are the average directions at cells i and j, respectively.

[0170] Furthermore, the above step S3.1 includes the following:

[0171] First, extract the edge information m corresponding to the region of interest carried by each trajectory v ;

[0172] Then, based on the entry edge l v and the exit edge l entry in m exit , generate the entry goal g in (entrygoal) and the exit goal g out (exitgoal);

[0173] Finally, group the set F of vehicle trajectories based on the entry goal g in and the exit goal g out . The vehicle trajectories within a group have the same entry goal g in and the same exit goal g out .

[0174] Furthermore, step S3.5 also includes the following: Each region of interest in the traffic flow field is designed as a separate grid.

[0175] The multi-channel traffic flow field G constitutes an elegant structure to encode human-like driving behaviors, maintained indexed by spatial location, and each region of interest polygon is maintained as a separate grid.

[0176] To achieve continuous update of traffic flow information, each region of interest will maintain a first-in-first-out (FIFO) queue, which replaces the older traffic flow with an earlier time when new traffic flow is observed. Through this mechanism, it is ensured that the traffic flow information reflects the latest road information.

[0177] In some embodiments, the resolution of the traffic flow field G can be set to 0.2 meters in detail, and each cell is associated with f i,j = (d, δ i , δ j ).

[0178] d is calculated by counting the number of vehicle trajectories overlapping with the cell, noting that the bounding box rather than the center point is used when calculating d.

[0179] For δ i and δ jThey are the average directions of all the trajectory lines overlapping with the unit.

[0180] The number of channels of the traffic flow field G depends on the total number of entry points of the clustering.

[0181] The storage of the traffic flow field G utilizes its sparse structure to save bandwidth.

[0182] The partial algorithms involved in step S3 are as follows:

[0183]

[0184]

[0185] Step S4: Perform path planning based on the traffic flow field.

[0186] Furthermore, step S4 includes:

[0187] Step S4.1: Load the traffic flow field;

[0188] Step S4.2: Enter the region of interest and query the traffic flow field data;

[0189] Step S4.3: Search for the optimal initial path based on the traffic density and the average direction of the vehicle flow;

[0190] Step S4.4: Establish a frenet coordinate system according to the optimal initial path;

[0191] Step S4.5: Based on the frenet coordinate system, sample the longitudinal nodes (stations) and the corresponding transverse clusters;

[0192] Step S4.6: Based on the graph formed by the longitudinal nodes and the transverse clusters, use the dynamic programming algorithm (DynamicSearch) to generate multiple potential candidate paths;

[0193] Step S4.7: Use non-maximum suppression to remove the duplicate paths in the candidate paths;

[0194] Step S4.8: Perform quadratic programming (QuadraticProgramming, QP) optimization on the candidate paths after removing the duplicate paths to smooth the paths.

[0195] Due to the discretization of the grid, the path is not smooth enough for the vehicle to trace. Therefore, local path smoothing based on quadratic programming (QuadraticProgramming, QP) is used to smooth the path.

[0196] In some embodiments, specifically for the optimal initial path, start from the entry point of each layer and use dynamic programming (DynamicSearch) to search for the path that best matches G z For the optimal initial path search, two weighted costs are designed, namely the density cost and the direction cost, which penalize paths that enter low-density areas or do not match the field direction.

[0197] The resolution of site sampling can be set to 3 meters. The lateral clustering strategy is similar to the entry point clustering. Another dynamic programming search is based on the graph formed by longitudinal sampling nodes (Stations) and lateral clusters (Clusters), where G z is used for cost evaluation. When performing non-maximum suppression (nonmaximumsuppress), first sort the candidate paths according to the costs output by dynamic programming, and then iterate starting from the path with the lowest cost. The suppression criterion is based on the percentage of nodes with a minimum lateral distance greater than 2 meters from the previous path. If a path contains more than 20% of the nodes with significant lateral differences, then this path will be regarded as a new candidate path.

[0198] Part of the algorithm involved in step S4 is as follows:

[0199]

[0200] The present invention has a complete method for processing open space traffic flow, including designing a sparse storage structure with low bandwidth and abstracting and clustering human-like vehicle flow trajectories.

[0201] The present invention provides a process for searching and optimizing vehicle executable paths from a traffic flow field, getting rid of the dependence on high-precision maps and real-time online detection.

[0202] The present invention provides a new idea for human-like planning, which can be used for the next stage of imitation learning training and road topology reasoning.

[0203] An embodiment of the present invention discloses a vehicle path planning device based on traffic flow, including:

[0204] An acquisition module, configured to acquire traffic flow from several vehicle trajectories generated by an in-vehicle tracking module;

[0205] A preprocessing module, configured to preprocess the traffic flow;

[0206] A conversion module, configured to convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure;

[0207] A path planning module, configured to perform path planning based on the traffic flow field.

[0208] Furthermore, the preprocessing module includes:

[0209] A primary filtering unit for filtering out unqualified vehicle trajectories in the traffic flow based on one or more of the tracked life cycle, the total travel distance of the trajectory, the presence of abnormal drift, and whether there is a collision with an obstacle;

[0210] A secondary filtering unit for further filtering the traffic flow to screen out vehicle trajectories with clear entry and exit points relative to the region of interest, where each vehicle trajectory carries edge information corresponding to the region of interest, obtaining

[0211] F = b v , m v |v ∈ V;

[0212] m v = (l entry , l exit , p start );

[0213] Where: F is the set of vehicle trajectories;

[0214] b v is the path of vehicle v at all tracked timestamps;

[0215] V is the set of vehicle tracking IDs;

[0216] m v is the information carried by the vehicle trajectory;

[0217] l entry and l exit are respectively the entry edge and the exit edge corresponding to the region of interest (i.e., the edges intersecting the traffic flow trajectory at the intersection);

[0218] p start is the intersection point of l entry .

[0219] Furthermore, the conversion module includes:

[0220] A first grouping unit for grouping the vehicle trajectories in the traffic flow according to the edge information corresponding to the region of interest carried by each vehicle trajectory, and the vehicle trajectories within the group have the same entry and exit;

[0221] A clustering unit for obtaining multiple entry points p e , p e through clustering for each group of vehicle trajectories, which is the channel-level identifier for each vehicle trajectory;

[0222] A second grouping unit for grouping according to the entry point p e to obtain the traffic flow corresponding to different entry points

[0223] Traffic flow field construction unit, used to construct a traffic flow field with a multi-channel structure, and each channel is the traffic flow from a specific entry point

[0224] Grid construction unit, used to construct a grid for each channel of the traffic flow field, and each cell of the grid stores the information f i,j =(d, δ i , δ j );

[0225] Where: d represents the traffic flow density, that is, the number of vehicles passing through this cell;

[0226] δ i , δ j are respectively the average directions at cells i and j.

[0227] As a preferred solution, the first grouping unit is used to execute the following methods, including:

[0228] First, extract the edge information m corresponding to the region of interest carried by each trajectory v ;

[0229] Then, based on the entry edge l v , exit edge l entry in m exit , generate the entry target g in and the exit target g out ;

[0230] Finally, group the set F of vehicle trajectories based on the entry target g in and the exit target g out , and the vehicle trajectories within the group have the same entry target g in and the exit target g out .

[0231] Furthermore, the grid construction unit is also used to execute the following content: Each region of interest in the traffic flow field is designed as a separate grid.

[0232] Furthermore, under the continuous update of traffic flow information, each region of interest in the traffic flow field is updated based on the first-in-first-out queue mechanism.

[0233] Furthermore, the path planning module includes:

[0234] Loading unit, used to load the traffic flow field;

[0235] Query unit, used to query the traffic flow field data by entering the region of interest;

[0236] Initial path search unit, used to search for the optimal initial path based on the traffic flow density and the average direction of the vehicle flow;

[0237] A coordinate system establishment unit, configured to establish a frenet coordinate system according to the optimal initial path;

[0238] A sampling unit, configured to sample longitudinal nodes and corresponding transverse clusters based on the frenet coordinate system;

[0239] A dynamic programming unit, configured to generate multiple potential candidate paths by using a dynamic programming algorithm based on a graph formed by longitudinal nodes and transverse clusters;

[0240] A duplicate path removal unit, configured to remove duplicate paths in the candidate paths;

[0241] A path smoothing unit, configured to perform quadratic programming optimization on the candidate paths after removing duplicate paths to smooth the paths.

[0242] The specific content of the vehicle path planning device based on traffic flow in this embodiment is similar to the content of the vehicle path planning method based on traffic flow disclosed in the above embodiment, and will not be elaborated here.

[0243] In addition, an embodiment of the present invention also discloses a storage medium, which stores one or more computer-readable programs, and one or more programs include instructions, and the instructions are adapted to be loaded and executed by the memory to perform any one of the above vehicle path planning methods.

[0244] A vehicle path planning method, device and storage medium of the present invention have the following beneficial effects:

[0245] First, for extra-large and irregular intersections, using a traffic flow field can achieve human-like path planning without using a high-definition map, greatly reducing the annotation cost.

[0246] Second, since its incremental update adopts a first-in, first-out traffic flow update strategy, it has good adaptability to change scenarios such as road construction and road routing changes, enabling the intelligent driving algorithm to update the planning result in a timely manner according to environmental changes.

[0247] Third, a traffic flow field can be used to detect potential map errors and timely feedback to update the high-definition map.

[0248] In the traditional high-precision map maintenance process, it is difficult to quickly locate these road construction and diversion events. However, using traffic flow information updates can identify these events. By comparing the traffic flow results with the high-definition map, mismatches can be easily identified, and a high-precision map update signal can be fed back using the traffic flow field.

[0249] The present invention extracts human-like driving paths from the traffic flow field, and can generate guiding paths even in large open spaces with few road markings.

[0250] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the methods and apparatuses of the present invention, or certain aspects or portions of the methods and apparatuses of the present invention, may take the form of program code (i.e., instructions) embedded in a tangible medium, such as a floppy disk, CD-ROM, hard disk drive, or any other machine-readable storage medium, wherein when the program is loaded into and executed by a machine such as a computer, the machine becomes an apparatus for practicing the present invention.

[0251] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have various changes and improvements, and these changes and improvements fall within the scope of the present invention as claimed. The scope of the present invention as claimed is defined by the appended claims and their equivalents.

Claims

1. A vehicle path planning method based on traffic flow, characterized in that: include: Step S1: obtaining traffic flow from a number of vehicle trajectories generated by a vehicle tracking module; Step S2: pre-processing the traffic flow; Step S3: converting the preprocessed traffic flow into a traffic flow field with a multi-channel structure; Step S4: Perform path planning based on the traffic flow field.

2. The vehicle path planning method according to claim 1, characterized in that: The step S2 comprises: Step S2.1: filtering unqualified vehicle trajectories in the traffic flow based on one or more factors including the tracking life cycle, the total travel distance of the trajectory, whether there is abnormal drift, and whether there is collision with obstacles; Step S2.2: Further filter the traffic flow to select vehicle trajectories with clear entry and exit points relative to the area of ​​interest. Each vehicle trajectory carries the edge information corresponding to the area of ​​interest. F=(b v ,m v )|v∈V; m v =(l entry ,l exit ,p start ); Where: F is the set of vehicle trajectories; b v is the path of vehicle v at all timestamps in the tracking; V is the set of vehicle tracking IDs; m v Carrying information for vehicle trajectories; l entry and l exit They are the entrance edge and exit edge corresponding to the region of interest respectively; p start for l entry The intersection of .

3. The vehicle path planning method according to claim 2, characterized in that: The step S3 comprises: Step S3.1: grouping the vehicle trajectories of the traffic flow according to the edge information corresponding to the region of interest carried by each vehicle trajectory, wherein the vehicle trajectories in the group have the same entrance and exit; Step S3.2: Obtain multiple entry points p for each group of vehicle trajectories by clustering e , p e is the channel-level identifier of each vehicle trajectory; Step S3.3: According to the entry point p e Group and obtain the flow F corresponding to different entry points <gin,gout,pe> ; Step S3.4: Construct a traffic flow field with a multi-channel structure, where each channel is the flow F from a specific entry point <gin,gout,pe> ; Step S3.5: For each channel of the traffic flow field, a grid is constructed, and each cell of the grid stores information f i,j =(d,δ i , δ j ); Where: d represents the traffic density, that is, the number of vehicles passing through the cell; δ i , δ j The average direction of cells i and j respectively.

4. The vehicle path planning method according to claim 3, characterized in that: The step S3.1 comprises: First, extract the edge information m corresponding to the region of interest carried by each trajectory v ; Then, according to m v Middle entrance edge entry , exit edge exit , generating the entry target g in and export target g out ; Finally, based on the entry target g in and export target g out Group the set F of vehicle trajectories, and the vehicle trajectories in the group have the same entry target g in and export target g out .

5. The vehicle path planning method according to claim 3, characterized in that: The step S3.5 also includes the following content: each region of interest in the traffic flow field is designed as a separate grid.

6. The vehicle path planning method according to claim 5, characterized in that: With the continuous updating of traffic flow information, each region of interest in the traffic flow field is updated based on a first-in-first-out queue mechanism.

7. The vehicle path planning method according to claim 3, characterized in that: The step S4 comprises: Step S4.1: Loading the traffic flow field; Step S4.2: Enter the area of ​​interest and query the traffic flow field data; Step S4.3: searching for the optimal initial path based on traffic density and average direction of traffic flow; Step S4.4: Establish the frenet coordinate system according to the optimal initial path; Step S4.5: based on the frenet coordinate system, sampling the vertical nodes and the horizontal clusters corresponding to the vertical nodes; Step S4.6: Based on the graph formed by the vertical nodes and horizontal clusters, a plurality of potential candidate paths are generated using a dynamic programming algorithm; Step S4.7: removing duplicate paths from candidate paths; Step S4.8: Perform secondary programming optimization on the candidate paths after removing duplicate paths to smooth the paths.

8. A vehicle path planning device based on traffic flow, characterized in that: include: An acquisition module, used for acquiring traffic flow from a number of vehicle trajectories generated by the vehicle tracking module; A preprocessing module, used for preprocessing traffic flow; A conversion module, used to convert the preprocessed traffic flow into a traffic flow field with a multi-channel structure; The path planning module is used to plan the path based on the traffic flow field.

9. The vehicle path planning device according to claim 8, characterized in that: The preprocessing module comprises: A preliminary filtering unit, for filtering unqualified vehicle trajectories in the traffic flow based on one or more factors of the tracking life cycle, the total travel distance of the trajectory, whether there is abnormal drift, and whether there is a collision with an obstacle; The secondary filtering unit is used to further filter the traffic flow and select the vehicle trajectories with clear entry and exit points relative to the area of ​​interest. Each vehicle trajectory carries the edge information corresponding to the area of ​​interest. F=(b v ,m v )|v∈V; m v =(l entry ,l exit ,p start ); Where: F is the set of vehicle trajectories; b v is the path of vehicle v at all timestamps in the tracking; V is the set of vehicle tracking IDs; m v Carrying information for vehicle trajectories; l entry and l exit They are the entrance edge and exit edge corresponding to the region of interest respectively; p start for l entry The intersection of .

10. The vehicle path planning device according to claim 9, characterized in that: The conversion module comprises: A first grouping unit is used to group the vehicle trajectories of the traffic flow according to the edge information corresponding to the area of ​​interest carried by each vehicle trajectory, and the vehicle trajectories in the group have the same entrance and exit; The clustering unit is used to obtain multiple entry points p for each group of vehicle trajectories through clustering e , p e is the channel-level identifier of each vehicle trajectory; The second grouping unit is used to group the e Group and obtain the flow F corresponding to different entry points <gin,gout,pe> ; Traffic flow field construction unit, used to construct a traffic flow field with a multi-channel structure, each channel is the flow F from a specific entry point <gin,gout,pe> ; The grid construction unit is used to construct a grid for each channel of the traffic flow field. Each cell of the grid stores information f i,j =(d,δ i , δ j ); Where: d represents the traffic density, that is, the number of vehicles passing through the cell; δ i , δ j The average direction of cells i and j respectively.

11. The vehicle path planning device according to claim 10, characterized in that: The first grouping unit is used to perform the following method, including: First, extract the edge information m corresponding to the region of interest carried by each trajectory v ; Then, according to m v Middle entrance edge entry , exit edge exit , generating the entry target g in and export target g out ; Finally, based on the entry target g in and export target g out Group the set F of vehicle trajectories, and the vehicle trajectories in the group have the same entry target g in and export target g out .

12. The vehicle path planning device according to claim 10, characterized in that: The grid construction unit is further used to perform the following: each region of interest in the traffic flow field is designed as a separate grid.

13. The vehicle path planning device according to claim 12, characterized in that: With the continuous updating of traffic flow information, each region of interest in the traffic flow field is updated based on a first-in-first-out queue mechanism.

14. The vehicle path planning device according to claim 10, characterized in that: The path planning module includes: A loading unit is used to load the traffic flow field; A query unit, used to enter an area of ​​interest and query traffic flow field data; An initial path search unit, used to search for an optimal initial path based on traffic density and average direction of traffic flow; A coordinate system establishment unit, used to establish a frenet coordinate system according to the optimal initial path; A sampling unit, used for sampling the longitudinal nodes and the lateral clusters corresponding to the longitudinal nodes based on the Frenet coordinate system; A dynamic programming unit, used to generate multiple potential candidate paths based on a graph formed by vertical nodes and horizontal clusters using a dynamic programming algorithm; A duplicate path removal unit, used for removing duplicate paths from candidate paths; The path smoothing unit is used to perform secondary planning optimization on the candidate paths after removing duplicate paths to smooth the paths.

15. A storage medium, characterized in that The storage medium stores one or more computer-readable programs, and the one or more programs include instructions, and the instructions are suitable for being loaded by the memory and executing the vehicle path planning method described in any one of claims 1-7.

Images