Intention recognition method, vehicle, electronic device, storage medium and program product

Abstract

The invention provides an intention recognition method, a vehicle, electronic equipment, a storage medium and a program product, and relates to the technical field of vehicles. According to the intention recognition method, map data is detected in response to a target vehicle, available state judgment is carried out on the map data, intention type misrecognition caused by incomplete map data or insufficient precision is avoided, and when the map data is in an unavailable state, intention type misrecognition is carried out. The first current scene category is determined according to the current environment information of the target vehicle and the vehicle state information, misrecognition of the intention type caused by insufficient scene adaptation is avoided, the intention of the target object is further recognized according to the first current scene category and the target sensing motion feature based on the preset rule table, and the target object intention recognition efficiency is improved. The dependence on map data is reduced, and the accuracy and robustness of intention recognition are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to an intention recognition method, a vehicle, an electronic device, a storage medium and a program product. BACKGROUND

[0002] In the application of autonomous driving (AD) and advanced driver assistance systems (ADAS), the intersection is one of the most frequent and risky scenarios for vehicles to interact with vulnerable road users (VRU) such as pedestrians and cyclists. Intersections usually contain complex traffic elements, such as lane boundaries, pedestrian crossings, traffic lights, construction areas, etc., and there are multiple dynamic participants (such as vehicles, pedestrians, cyclists) working together. The behavior intention of VRU (such as whether to cross the road, whether to slow down and stop, or whether to change direction) directly affects the decision logic of the autonomous driving system (such as braking, lane changing, or path planning).

[0003] In related technologies, the recognition of VRU behavior intention usually relies on static analysis based on map topology. Specifically, by matching the road topology information (such as lane boundaries, pedestrian crossings, curbs, traffic lights, etc.) in the map data with the position information of the VRU, the intention of the vulnerable road user is inferred, such as whether it is possible to cross the road or stop moving.

[0004] However, the above-mentioned method of identifying VRU intention based on map topology has strong dependence on the map, and in some scenarios where the map data is incomplete or not updated in real time, there is a problem of low accuracy in recognizing the intention of the VRU. SUMMARY

[0005] The present application provides an intention recognition method, a vehicle, an electronic device, a storage medium and a program product to solve the problem of low accuracy in recognizing the intention of the VRU in related technologies based on map topology to identify the intention of the VRU, which has strong dependence on the map, and in some scenarios where the map data is incomplete or not updated in real time.

[0006] In a first aspect, the present application provides an intention recognition method, comprising: in response to a target vehicle detecting map data, determining whether the map data is in an available state; if the map data is in an unavailable state, determining a first current scene category according to current environmental information of the target vehicle and self-vehicle state information; determining a target perception motion feature of a target object according to the self-vehicle state information and object perception information of the target vehicle; and identifying the intention of the target object based on a preset rule table, according to the first current scene category and the target perception motion feature.

[0007] In the embodiment of the application, in response to the target vehicle detecting the map data, the intention type misrecognition caused by incomplete or insufficient accuracy of the map data is avoided by performing the available state determination on the map data, and when the map data is in an unavailable state, the intention type misrecognition caused by insufficient scene adaptation is avoided by determining the first current scene category according to the current environment information of the target vehicle and the ego vehicle state information, and further, the intention of the target object is recognized according to the first current scene category and the target perception motion feature based on the preset rule table, so as to reduce the dependence on the map data and improve the accuracy and robustness of intention recognition.

[0008] In a possible implementation, the ego vehicle state information includes ego vehicle position coordinates and ego vehicle orientation, and the first current scene category is determined according to the current environment information of the target vehicle and the ego vehicle state information, including: performing regional division on the current driving environment of the target vehicle according to the ego vehicle position coordinates and the current environment information to obtain a target region corresponding to the target vehicle; and determining the first current scene category according to the regional road feature of the target region and the ego vehicle orientation.

[0009] In the embodiment of the application, the current driving environment of the target vehicle is regionally divided according to the ego vehicle position coordinates and the current environment information to obtain a target region corresponding to the target vehicle, and further, the first current scene category is determined according to the regional road feature of the target region and the ego vehicle orientation, so as to realize the explicit classification of the current scene in which the target vehicle is located, to provide structured scene background information for intention recognition of the target object, and to improve the intention recognition accuracy.

[0010] In a possible implementation, the object perception information includes at least one first object, and state information and category identification information of each first object, and the target perception motion feature of the target object is determined according to the ego vehicle state information and the object perception information of the target vehicle, including: screening the at least one first object according to the ego vehicle state information and the state information and category identification information of each first object to obtain the target object; and calculating the target perception motion feature of the target object according to the state information of the target object and the ego vehicle state information.

[0011] In the embodiment of the application, the target object is obtained by screening the at least one first object according to the ego vehicle state information and the state information and category identification information of each first object, and further, the target perception motion feature of the target object is calculated according to the state information of the target object and the ego vehicle state information, so as to reduce the calculation complexity and focus the decision process based on the target vehicle on the most critical target entity in the scene.

[0012] In a possible implementation, the intention of the target object is identified according to the first current scene category and the target perception motion feature based on the preset rule table, including: generating structured data according to the first current scene category and the target perception motion feature; matching the structured data in the preset rule table, and outputting the intention type of the target object and the confidence of the intention type.

[0013] In a possible implementation, the intention recognition method further includes: if the map data is in an available state, determining a second current scene category according to the map data and the ego vehicle state information; identifying the intention of the target object according to the second current scene category; or identifying the intention of the target object according to the second current scene category and the target perception motion feature.

[0014] In a possible implementation, the second current scene category is determined according to the map data and the ego vehicle state information, including: performing coordinate conversion on the latitude and longitude of the map data to obtain global plane coordinates corresponding to the map data; performing coordinate conversion on the global plane coordinates to obtain a local plane coordinate system corresponding to the target vehicle; and determining the second current scene category according to the local plane coordinate and the ego vehicle state information.

[0015] In a possible implementation, the preset rule table is constructed by: obtaining a sample data set, the sample data set including road topology data, ego vehicle data, a plurality of different second objects, and initial motion trajectories and perception data of each second object; performing true value labeling on the initial motion trajectory of each second object in the plurality of different second objects to obtain a target motion trajectory of the second object, the target motion trajectory carrying a true value intention type of the second object; extracting perception motion features and topology features of the second object according to the road topology data, the ego vehicle data, and the perception data of the second object; grouping based on the true value intention type, identifying feature combination rules of different true value intention types according to the true value intention type, the topology features, and the perception motion features of each second object; and constructing the preset rule table according to the feature combination rules.

[0016] Embodiments of the present application construct a preset rule table by adopting a comprehensive data-driven process, and identify the intention of the target object based on the preset rule table, so as to ensure that the intention recognition method has explainability, data-driven effectiveness, and real-time applicability.

[0017] In a second aspect, the present application provides an intention recognition device, including:

[0018] The first determination module is configured to determine whether the map data is in an available state in response to the target vehicle detecting the map data.

[0019] The second determining module is configured to determine a first current scene category according to the current environment information of the target vehicle and the ego vehicle state information when the map data is in an unavailable state.

[0020] The third determining module is configured to determine a target perception motion feature of the target object according to the ego vehicle state information and the object perception information of the target vehicle.

[0021] The identifying module is configured to identify an intention of the target object according to the first current scene category and the target perception motion feature based on a preset rule table.

[0022] In a possible implementation, the ego vehicle state information includes ego vehicle position coordinates and ego vehicle orientation, and the second determining module is specifically configured to: divide a current driving environment of the target vehicle into target regions according to the ego vehicle position coordinates and the current environment information, to obtain a target region corresponding to the target vehicle; and determine the first current scene category according to a regional road feature of the target region and the ego vehicle orientation.

[0023] In a possible implementation, the object perception information includes at least one first object, and state information and category identification information of each first object, and the third determining module is specifically configured to: filter the at least one first object according to the ego vehicle state information and the state information and the category identification information of each first object, to obtain the target object; and calculate the target perception motion feature of the target object according to the state information of the target object and the ego vehicle state information.

[0024] In a possible implementation, the identifying module is specifically configured to: generate structured data according to the first current scene category and the target perception motion feature; and match the structured data in the preset rule table, to output an intention type of the target object and a confidence degree of the intention type.

[0025] In a possible implementation, the second determining module is further configured to: determine a second current scene category according to the map data and the ego vehicle state information when the map data is in an available state; identify the intention of the target object according to the second current scene category; or identify the intention of the target object according to the second current scene category and the target perception motion feature.

[0026] In a possible implementation, the second determining module is further configured to: perform coordinate conversion on longitude and latitude of the map data, to obtain global plane coordinates corresponding to the map data; perform coordinate conversion on the global plane coordinates, to obtain a local plane coordinate system corresponding to the target vehicle; and determine the second current scene category according to the local plane coordinate and the ego vehicle state information.

[0027] In a possible implementation, the preset rule table is constructed by: obtaining a sample data set, the sample data set including road topology data, ego vehicle data, a plurality of different second objects, and initial motion trajectories and perception data of each second object; for each second object in the plurality of different second objects, performing true value labeling on the initial motion trajectory of the second object to obtain a target motion trajectory of the second object, the target motion trajectory carrying a true value intention type of the second object; extracting perception motion features and topology features of the second object according to the road topology data, the ego vehicle data, and the perception data of the second object; grouping based on the true value intention type, identifying feature combination rules of different true value intention types according to the true value intention type, the topology features, and the perception motion features of each second object; and constructing the preset rule table according to the feature combination rules.

[0028] In a third aspect, the present application provides a vehicle, comprising: a vehicle body and an intention recognition system, the intention recognition system being configured to implement the intention recognition method provided in the first aspect.

[0029] In a fourth aspect, the present application provides an electronic device, comprising: a processor and a memory connected with the processor in communication; the memory stores computer execution instructions; and the processor executes the computer execution instructions stored in the memory to implement the intention recognition method provided in the first aspect.

[0030] In a fifth aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing computer execution instructions, the computer execution instructions being configured to implement the intention recognition method provided in the first aspect when executed by a processor.

[0031] In a sixth aspect, the present application provides a computer program product, comprising: a computer program, the computer program being configured to implement the intention recognition method provided in the first aspect when executed by a processor.

[0032] The application provides an intention recognition method, a vehicle, an electronic device, a storage medium and a program product. The intention recognition method comprises the following steps: in response to the target vehicle detecting the map data, determining whether the map data is in an available state; when the map data is in an unavailable state, determining a first current scene category according to current environment information and self-vehicle state information of the target vehicle; further determining a target perception motion feature of a target object according to the self-vehicle state information and object perception information of the target vehicle; and then, based on a preset rule table, recognizing the intention of the target object according to the first current scene category and the target perception motion feature. According to the application, in response to the target vehicle detecting the map data, the available state of the map data is determined, so that the intention type misrecognition caused by incomplete or insufficient accuracy of the map data is avoided. When the map data is in the unavailable state, the first current scene category is determined according to the current environment information and the self-vehicle state information of the target vehicle, so that the intention type misrecognition caused by insufficient scene adaptation is avoided. Further, based on the preset rule table, the intention of the target object is recognized according to the first current scene category and the target perception motion feature, so that the dependence on the map data is reduced, and the accuracy and robustness of intention recognition are improved. BRIEF DESCRIPTION OF DRAWINGS

[0033] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0034] Figure 1 An architecture diagram of an intention recognition system provided by an embodiment of the application;

[0035] Figure 2 A flowchart of an intention recognition method provided by an embodiment of the application Figure 1 ;

[0036] Figure 3 A flowchart of an intention recognition method provided by an embodiment of the application Figure 2 ;

[0037] Figure 4 A flowchart of constructing a preset rule table provided by an embodiment of the application;

[0038] Figure 5 A structural diagram of an intention recognition device provided by an embodiment of the application;

[0039] Figure 6 A structural diagram of an electronic device provided by an embodiment of the application.

[0040] The specific embodiments of the application have been shown by the above drawings, and will be described in more detail hereinafter. These drawings and detailed description are not intended to limit the scope of the concept of the application in any way, but to illustrate the concept of the application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0041] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The following description is made with reference to the accompanying drawings in which like reference numerals refer to like elements, unless the context of use indicates otherwise. The following description of exemplary embodiments is not representative of all embodiments consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0042] In the related art, the recognition of the intention of the VRU behavior usually relies on static analysis based on map topology. Specifically, by matching the road topology information (such as lane boundary, pedestrian crossing, curb, traffic signal, etc.) in the map data with the position information of the VRU, the intention of the vulnerable road user is inferred, such as whether it is possible to cross the road or stop moving. However, this intention recognition method has the following limitations: 1) when the map data is incomplete (such as not covering all intersections), not updated in time (such as the map data is not updated regularly) or insufficient in accuracy, there is a problem of low accuracy of intention recognition; 2) strong dependence on map data, unable to realize intention recognition when lacking map data, unable to cope with unstructured or unknown intersection scenarios, and insufficient generalization ability; 3) when the environment changes dynamically (such as there are construction areas, temporary signal lights or blurred markings, etc.), resulting in inconsistency between the actual map topology and the static topology of the map, there is a problem of low accuracy of intention recognition.

[0043] Based on the technical problems existing in the related art, the embodiments of the present application avoid misrecognition of the intention type caused by incomplete or insufficient accuracy of the map data by determining the available state of the map data when the vehicle detects the map data, and avoid misrecognition of the intention type caused by insufficient scene adaptation by determining the current scene category according to the current environmental information and the ego state information of the vehicle when the map data is in an unavailable state. Further, the intention of the target object is recognized based on the preset rule table, according to the current scene category and the perceived motion feature, reducing the dependence on the map data and improving the accuracy and robustness of the intention recognition.

[0044] The application scenarios of the embodiments of the present application will be described first as follows.

[0045] The intent recognition method provided in this application is applicable to intersection scenarios in autonomous driving systems (AD) and advanced driver assistance systems (ADAS). Specifically, it is applicable to recognizing the intent of VRUs such as pedestrians and cyclists in intersection scenarios by combining topological features and perceived motion features.

[0046] Figure 1 This is an architecture diagram of the intent recognition system provided in an embodiment of this application. Figure 1 As shown, the intent recognition system includes a perception layer, a processing layer, and a decision layer.

[0047] The perception layer is used to detect map data and collect real-time environmental information, vehicle status information, and object perception information of the target vehicle based on sensors integrated on the target vehicle.

[0048] The processing layer includes a topology feature processing module, a motion perception feature processing module, a dataset, a rule-based intent selector, and a recognition module. The topology feature processing module performs current scene classification of the target vehicle; the motion perception feature processing module filters target objects based on an object decision-maker or filter, and calculates the target object's motion perception features; the dataset stores structured data generated based on a first current scene category or a second scene category and the target motion perception features; the rule-based intent selector infers conditional rules and constructs a preset rule table based on these rules; and the recognition module identifies the target object's intent based on the structured data and the preset rule table.

[0049] The decision layer is used to output the identified intent type to decision modules such as path planning and braking control.

[0050] Understandable, Figure 1 The topological feature processing module shown can be the second determining module in the intent recognition device of this application embodiment (e.g., Figure 5 (As shown in the diagram), all are used to perform current scene classification for the target vehicle. Figure 1 The motion processing module shown can be the third determining module in the intent recognition device of this application embodiment (e.g., Figure 5 (See illustration), all are used to filter target objects based on object decision-makers or filters, and to calculate the target perception motion features of target objects.

[0051] The following is based on Figure 1 The intent recognition system shown is the execution subject, and the specific implementation of the intent recognition method provided in the embodiments of this application will be described in detail.

[0052] Figure 2 Flowchart of the intent recognition method provided in the embodiments of this application Figure 1 .likeFigure 2 As shown, a specific implementation of this intent recognition method may include the following steps:

[0053] S201, in response to the target vehicle detecting map data, determines whether the map data is available.

[0054] For example, the map data can be high-precision electronic map data, including basic geographic data and road topology information. Among them, the basic geographic data includes, but is not limited to, the geometric location and attribute information of lane boundaries, sidewalks, crosswalks, curbs, traffic lights, lane center lines, and stop lines, etc., and the road topology information includes, but is not limited to, the connection relationship between lanes, the topology of intersection traffic directions, the connection relationship between crosswalks and adjacent lanes, and the correspondence between stop lines and lanes.

[0055] In this step, one possible way to determine whether the map data is available is to compare it with the map data based on the real-time environmental information (including local road features and local road topology information) collected by the target vehicle, determine the completeness and timeliness of the map data, and determine that the map data is available when it is determined that the map data is complete and meets the timeliness of the update. If the map data is incomplete or does not meet the timeliness of the update, it is determined that the map data is unavailable.

[0056] For example, one possible way to determine the integrity of map data is to determine whether the map data can cover the current environmental information collected by the target vehicle. If the map data can cover the current environmental information of the target vehicle, the map data is determined to be complete; if the map data cannot cover the current environmental information of the target vehicle, the map data is determined to be incomplete.

[0057] For example, one possible implementation for determining the timeliness of map data updates is to determine whether the map data version is the latest version or whether the map data is consistent with the current road conditions collected by the target vehicle (such as whether there are any newly added construction areas). If the map data version is the latest version or the map data is consistent with the current road conditions collected by the target vehicle, the map data is determined to meet the timeliness requirement. If the map data version is not the latest version or the map data is inconsistent with the current road conditions collected by the target vehicle, the map data is determined not to meet the timeliness requirement.

[0058] For example, in some embodiments, the determination of the availability of map data may also include determining whether the accuracy of the map data meets the required standards.

[0059] This application does not limit the criteria for determining whether map data is available; the criteria can be determined based on actual application needs.

[0060] S202, when map data is unavailable, determine the first current scene category based on the target vehicle's current environmental information and vehicle status information.

[0061] For example, the target vehicle can be a vehicle that integrates an autonomous driving system or an advanced driver assistance system.

[0062] For example, the current environmental information may include current road information and current road topology information. The current road information includes, but is not limited to, the geometric positions and attributes of lane boundaries, sidewalks, crosswalks, curbs, traffic lights, lane centerlines, and stop lines in the target vehicle's current surrounding environment. The current road topology information includes, but is not limited to, the connectivity between lanes, the intersection's traffic direction topology, the connection between crosswalks and adjacent lanes, and the correspondence between stop lines and lanes in the target vehicle's current surrounding environment. The target vehicle's current surrounding environment can be the range that the sensors integrated on the target vehicle can perceive.

[0063] It is understandable that the basic geographic data and road topology information contained in the map data are global information, while the current road information and current road topology information contained in the current environment information are local information about the current surrounding environment of the target vehicle.

[0064] It is understandable that the self-vehicle status information is the same as the target vehicle's status information.

[0065] For example, the vehicle status information includes, but is not limited to, the target vehicle's speed, acceleration, orientation, position coordinates, and geographic shape information within a specified spatial range (bbox). The geographic shape information within the specified spatial range (bbox) may include the parallelism between the target vehicle's body outline and the lane boundary, and the fit between the vehicle's width and the lane width.

[0066] For example, the first current scene category can be the scene in which the target vehicle is located relative to the intersection.

[0067] For example, the first current scenario category includes, but is not limited to, approaching an intersection, entering an intersection, being inside an intersection, exiting an intersection, leaving an intersection, turning right, and turning left.

[0068] For example, the target vehicle's current environmental information and vehicle status information can be obtained in real time through sensors integrated on the target vehicle, such as cameras, lidar, and millimeter-wave radar.

[0069] S203, based on the vehicle's status information and the target vehicle's object perception information, determine the target object's target perception motion characteristics.

[0070] For example, object perception information can be the state information and category identification information of dynamic and static objects detected by sensors integrated on the target vehicle in the current driving environment. Among them, the state information includes, but is not limited to, position, speed, acceleration, and orientation, and the category identification information includes, but is not limited to, pedestrians, cyclists, motor vehicles, electric vehicles, construction fences, and roadblocks.

[0071] For example, the target object can be a VRU (such as a pedestrian or cyclist) that has a strong spatial correlation with the target vehicle. The spatial correlation can be determined based on the relative distance between the target vehicle and the dynamic object. For example, the smaller the relative distance, the stronger the spatial correlation, and the larger the relative distance, the weaker the spatial correlation.

[0072] For example, target perception motion characteristics include, but are not limited to, the relative distance and relative position between the target vehicle and the target object, the distance type (near, medium, or long distance), the motion trend (the target object is approaching the target vehicle, the target object is moving away from the target vehicle, or the target object is maintaining a relative distance from the target vehicle), the time to collision (TTC), and the collision risk level. The collision time can be defined as the estimated time for a collision to occur, assuming the target object and the target vehicle maintain their current state of motion at their current relative speed and relative distance.

[0073] It should be noted that there is no difference in the order of steps S202 and S203.

[0074] S204, based on a preset rule table, identify the target object's intent according to the first current scene category and the target's perceived motion characteristics.

[0075] For example, a preset rule table can be used to represent the mapping relationship between a specific combination of perceptual motion features and topological features strongly related to a specific intent type and the specific intent type, or to represent the mapping relationship between topological features strongly related to a specific intent type and the specific intent type, or to represent the mapping relationship between perceptual motion features strongly related to a specific intent type and the specific intent type, etc. The topological features may include scene categories.

[0076] The types of preset rule tables in this application embodiment are not limited; they can be determined according to actual application requirements.

[0077] For example, the preset rule table also includes a confidence level for each intent type. The confidence level reflects the accuracy of the intent type.

[0078] For example, the types of intent include, but are not limited to, crossing the road and not crossing the road.

[0079] Table 1 is a preset rule table provided in the embodiments of this application for representing the mapping relationship between a specific combination of perceptual motion features and topological features strongly related to a specific intention type and a specific intention type.

[0080] Table 1

[0081]

[0082] As shown in Table 1, in Scenario 1, when the object type is a pedestrian, the scene type is at an intersection, the distance type in the perceived motion features is near, the relative position is the object being to the right of the target vehicle, the motion trend is approaching, the collision time is 1.2s, and the collision risk level is high, the intention type corresponding to the feature combination is crossing the road, and the confidence level corresponding to this intention type is X1; in Scenario 2, when the object type is a pedestrian, the scene type is at an intersection, the distance type in the perceived motion features is near, and the relative position is the object being to the right of the target vehicle... In scenario 3, when the object type is a pedestrian, the scene type is located at an intersection, the distance type of the perceived motion feature is close, the relative position is the object being to the right rear of the target vehicle, the motion trend is approaching, the collision time is infinite, and the collision risk level is low, the intention type of the feature combination is not to cross the road, and the confidence level of this intention type is X2.

[0083] Table 2 is a preset rule table provided in the embodiments of this application for representing the mapping relationship between topological features strongly related to a specific intent type and the specific intent type.

[0084] Table 2

[0085]

[0086] As shown in Table 2, in Scenario 1, when the object type is pedestrian and the scene type is located at an intersection, the corresponding intent type is crossing the road, and the confidence level of this intent type is X4.

[0087] It should be noted that the preset rule mapping tables shown in Tables 1 and 2 above are only examples. The embodiments of this application do not limit the feature combination form corresponding to the intent type in the preset rule table. The specific form can be determined according to the actual application requirements.

[0088] For example, the identified intent type of the target object can be used in the decision-making module of an advanced driver assistance system or autonomous driving system to achieve trajectory prediction or decision-making for the target vehicle.

[0089] In this embodiment, in response to the detection of map data by the target vehicle, the availability status of the map data is determined to avoid misidentification of intent type due to incomplete or insufficient map data. When the map data is unavailable, a first current scene category is determined based on the target vehicle's current environmental information and vehicle status information to avoid misidentification of intent type due to insufficient scene adaptation. Furthermore, based on a preset rule table, the intent of the target object is identified according to the first current scene category and the target's perceived motion characteristics, reducing the dependence on map data and improving the accuracy and robustness of intent recognition.

[0090] It is understandable that even when the target vehicle cannot detect map data (i.e., the target vehicle cannot obtain map data), intent recognition of the target object can still be achieved based on the target vehicle's current environmental information, vehicle status information, and object perception information. The specific implementation method is similar to that described above and will not be repeated here.

[0091] Optionally, the vehicle status information includes the vehicle's position coordinates and vehicle orientation. One possible implementation of step S202, which determines the first current scene category based on the target vehicle's current environment information and vehicle status information, is as follows: the target vehicle's current driving environment is divided into regions based on the vehicle's position coordinates and current environment information to obtain the target region corresponding to the target vehicle; the first current scene category is determined based on the regional road features of the target region and the vehicle's orientation.

[0092] For example, the vehicle's position coordinates, vehicle orientation, current environment information, and first current scene category are similar to those described above, and will not be repeated here.

[0093] For example, regional road features include, but are not limited to, the types of intersections (such as crossroads), pedestrian crossings, and straight road sections contained within the target area.

[0094] In this embodiment, one possible implementation is as follows: Using the vehicle's position coordinates as the center, and based on a preset size threshold, a target area is divided within the target vehicle's current driving environment, combined with current environmental information. Then, based on the regional road features of the target area and the vehicle's orientation, a first current scene category is determined. The preset size threshold can be dynamically adjusted based on the target vehicle's current driving conditions. This embodiment does not limit the preset size threshold; it can be determined according to actual application requirements.

[0095] In this embodiment, the current driving environment of the target vehicle is divided into regions based on the vehicle's position coordinates and current environmental information to obtain the target region corresponding to the target vehicle. Furthermore, based on the regional road characteristics of the target region and the vehicle's orientation, a first current scene category is determined, thereby achieving explicit classification of the scene in which the target vehicle is currently located. This provides structured scene background information for the intention recognition of the target object, improving the accuracy of intention recognition.

[0096] Optionally, the object-aware information includes at least one first object, as well as state information and category identification information for each first object.

[0097] For example, the first object can be a VRU and other vehicles besides the target vehicle, and the VRU can be a pedestrian or cyclist. Among them, cyclists can include bicycle riders and electric vehicle riders.

[0098] The status information and category identification information of each first object are similar to those described above, and will not be repeated here.

[0099] The following is combined with Figure 3 A detailed explanation is provided on a specific implementation method for step S203, which determines the target perception motion characteristics of the target object based on the vehicle's state information and the target vehicle's object perception information.

[0100] Figure 3 Flowchart of the intent recognition method provided in the embodiments of this application Figure 2 .like Figure 3 As shown, a specific implementation of this intent recognition method for determining the target perception motion characteristics of a target object based on the vehicle's state information and the target vehicle's object perception information may include the following steps:

[0101] S301, based on the vehicle status information and the status information and category identification information of each first object, at least one first object is filtered to obtain the target object.

[0102] For example, one possible implementation is as follows: First, at least one first object is filtered according to the category identification information of each first object to filter out non-VRU objects in at least one object. Then, based on the vehicle status information and the status information of each filtered first object, the perceived motion characteristics such as distance type between each filtered first object and the target vehicle are determined. Furthermore, based on the distance type between each first object and the target vehicle and the category identification information of each first object, the target object is further filtered from the filtered first objects.

[0103] For example, when the category identification information of the first object is pedestrian and the initial distance type is short distance or medium distance, the first object is determined to be the target object.

[0104] It is understandable that the corresponding perceptual motion characteristics of the target object may differ depending on the scenario. This application does not impose any restrictions on this, and the specific characteristics can be determined according to the actual application requirements.

[0105] It is understandable that the number of target objects obtained through filtering can be one or more.

[0106] S302, calculate the target perception motion characteristics of the target object based on the state information of the target object and the state information of the vehicle.

[0107] The motion characteristics of target perception are similar to those described above, and will not be repeated here.

[0108] For example, one possible implementation is as follows: Calculate the relative position of the target object and the target vehicle based on the vehicle's position coordinates and the target object's position coordinates; calculate the relative distance between the target object and the target vehicle based on the vehicle's position coordinates, the vehicle's geographic shape information within a specified spatial range (bbox), the target object's position coordinates, and the target object's geographic shape information within the specified spatial range (bbox); determine the distance type based on the relative distance; calculate the motion trend based on the vehicle's orientation, the target object's geographic shape information within the specified spatial range (bbox), and the target object's orientation; determine the distance type (short distance, medium distance, and long distance) based on the relative distance; calculate the collision time based on the vehicle's speed, the target object's speed, and the relative speed; and determine the collision risk level based on the collision time and the distance type.

[0109] In this embodiment, at least one first object is filtered based on the vehicle's state information and the state information and category identification information of each first object to obtain the target object. Furthermore, the target perception motion characteristics of the target object are calculated based on the state information of the target object and the vehicle's state information, thereby reducing computational complexity and enabling the decision-making process based on the target vehicle to focus on the most critical target entity in the scene.

[0110] Optionally, step S204 identifies the intent of the target object based on a preset rule table, according to the first current scene category and the target's perceived motion characteristics, including: generating structured data according to the first current scene category and the target's perceived motion characteristics; matching the structured data in the preset rule table; and outputting the intent type of the target object and the confidence level of the intent type.

[0111] In this embodiment, the preset rule table can be as shown in Table 1 above. The scene category corresponding to the topological feature can be determined based on the target vehicle's current environmental information and the vehicle's state information; its specific implementation is similar to that described above and will not be repeated here.

[0112] For example, structured data can be represented in dictionary form, including a first current scene category and target-aware motion features.

[0113] For example, structured data can be represented in the following ways:

[0114] Dictionary: {Target object number, [Target object type, Distance type, Relative position, Motion trend, Collision time, Collision risk level, Scene category...]}.

[0115] One possible implementation of matching structured data in a preset rule table and outputting the intent type of the target object and the confidence level of the intent type is as follows: match structured data in the preset rule table to obtain feature combinations that are consistent with the structured data, and determine the intent type corresponding to the feature combination as the intent type of the target object.

[0116] For example, when matching structured data in a preset rule table, if no matching feature combination is found, a prompt message is output indicating that the intent of the target object could not be recognized. This structured data is then recorded for analysis of the reasons for the intent recognition failure.

[0117] Optionally, the intent recognition method provided in this application embodiment further includes: if the map data is available, determining a second current scene category based on the map data and the vehicle status information; recognizing the intent of the target object based on the second current scene category; or recognizing the intent of the target object based on the second current scene category and the target's perceived motion characteristics.

[0118] Optionally, one possible way to determine the second current scene category based on map data and vehicle status information is as follows: perform coordinate transformation on the latitude and longitude of the map data to obtain the global plane coordinates corresponding to the map data; perform coordinate transformation on the global plane coordinates to obtain the local plane coordinate system corresponding to the target vehicle; and determine the second current scene category based on the local plane coordinates and vehicle status information.

[0119] For example, the file format of map data can be .osm format. This application does not limit the file format of map data.

[0120] For example, the local planar coordinate system can be a coordinate system with the target vehicle as the origin.

[0121] Understandably, by converting global planar coordinates into the local planar coordinate system corresponding to the target vehicle, the absolute position in geographic space is transformed into the relative position from the vehicle's perspective, so that the system can process local environmental information more efficiently and improve the classification accuracy of the second current scene category.

[0122] For example, one possible implementation of determining the second current scene category based on local planar coordinates and vehicle status information is as follows: taking the vehicle's position coordinates in the local planar coordinate system as the center, dividing the target area in the current driving environment of the target vehicle based on a preset size threshold and current environmental information, and determining the second current scene category based on the regional road features of the target area and the vehicle's orientation.

[0123] For example, another possible implementation of determining the second current scene category based on local planar coordinates and vehicle state information is as follows: taking the position coordinates of the vehicle in the local planar coordinate system as the center, and dividing the target area in the current driving environment of the target vehicle based on a preset size threshold, and determining the second current scene category based on the regional road features of the target area and the vehicle orientation.

[0124] For example, one possible way to identify the intent of the target object based on the second current scene category is to match the second current scene category and the type of the target object in the preset rule table shown in Table 2 above, and determine the intent type corresponding to the scene category and object type that are consistent with the second current scene category and the type of the target object as the intent of the target object.

[0125] For example, the specific implementation of identifying the target object's intent based on the second current scene category and the target's perceived motion characteristics is similar to the above, and will not be repeated here.

[0126] The following is combined with Figure 4 The construction method of the preset rule table provided in the embodiments of this application will be described in detail.

[0127] Figure 4 This is a schematic diagram illustrating the process of constructing a preset rule table as provided in an embodiment of this application. Figure 4 As shown, the construction method of this preset rule table specifically includes the following steps:

[0128] S401, Obtain a sample dataset, which includes road topology data, vehicle data, multiple different second objects, and the initial motion trajectory and perception data of each second object.

[0129] For example, the sample dataset can be a large-scale traffic scene dataset (covering pedestrians, cyclists, etc.) that includes various VRU interaction behaviors.

[0130] S402, for each of the multiple different second objects, perform truth labeling on the initial motion trajectory of the second object to obtain the target motion trajectory of the second object, which carries the truth intent type of the second object.

[0131] In one possible implementation, the initial motion trajectory of each second object is labeled with its true intention type (e.g., crossing the road, waiting in place, walking parallel, moving closer to the vehicle, etc.) to obtain the target motion trajectory of the second object.

[0132] S403: Based on road topology data, vehicle data, and perception data of the second object, extract the perception motion features and topology features of the second object.

[0133] The perception of motion features and topological features is similar to that described above, and will not be repeated here.

[0134] S404, based on truth intention type grouping, identifies the feature combination rules of different truth intention types according to the truth intention type, topological features and perceptual motion features of each second object.

[0135] One possible implementation involves statistically analyzing the distribution of these topological features and perceived motion features according to the truth intent category, identifying feature patterns with commonalities and distinctiveness. For example, pedestrians with a truth intent type of crossing the road typically exhibit a trend of "distance approaching" and a collision time value below a certain threshold, while pedestrians walking parallel to a vehicle maintain a relatively constant relative distance, and the collision time is considered infinite.

[0136] S405, Construct a preset rule table based on the feature combination rules.

[0137] In one possible implementation, "if-then" conditional rules are formulated based on feature combination patterns. Each conditional rule corresponds to a set of perceptual motion features and topological features strongly correlated with a specific intent type. Based on these conditional rules, an initial preset rule table is constructed. This initial preset rule table is applied to an independent validation dataset. By comparing the intent type predictions based on the initial preset rule table with the labeled ground truth intent types, the accuracy and confidence of each rule are evaluated. Based on rules that consistently maintain high accuracy and high confidence, a new preset rule table is constructed, while rules with weak relevance or unclear meaning are removed or further optimized. The confidence level is calculated based on statistical probability, rule matching degree, or the accuracy of the validation dataset.

[0138] For example, if (object type = pedestrian) and (distance type = near distance) and (movement trend = approaching) and (collision time) If the expression is ), then the corresponding intent type is crossing the road.

[0139] It is understood that the preset rule table provided in the embodiments of this application is not simply heuristic logic, but a concise and interpretable representation of the motion-intent relationship in real traffic scenario data that has been statistically verified.

[0140] In this embodiment, a preset rule table is constructed using a comprehensive data-driven process, and the intent of the target object is identified based on the preset rule table, ensuring that the intent identification method has interpretability, data-driven effectiveness, and real-time applicability.

[0141] In summary, the intent recognition method provided in this application has the following beneficial effects:

[0142] 1) By explicitly classifying the current scene of the target vehicle, structured scene background information is provided for the intention recognition of the target object, thereby improving the accuracy of intention recognition;

[0143] 2) Compared with related technologies that rely on map data for intent recognition and cannot perform intent recognition when there is no map data, the intent recognition method provided in this application embodiment can realize the intent recognition of the target object based solely on the target vehicle's perceived current environment information, vehicle status information, and object perception information, thereby improving the robustness of intent recognition.

[0144] 3) The intent recognition method provided in this application realizes multi-path fusion intent recognition by determining the availability status of map data, namely, the path with perceived motion features, the path with perceived motion features and topological features fused together, and the path with topological features.

[0145] 4) The intent recognition method provided in this application can be applied to urban intersections equipped with high-precision electronic maps, as well as to rural or unstructured intersections without map data. It can ensure robust recognition of VRU intent in both structured and unstructured traffic environments, thereby improving generalization ability.

[0146] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0147] Figure 5 This is a schematic diagram of the structure of the intent recognition device provided in an embodiment of this application. Figure 5 As shown, the intent recognition device 50 includes a first determination module 510, a second determination module 520, a third determination module 530, and an recognition module 540.

[0148] The first determining module 510 is used to determine whether the map data is available in response to the target vehicle detecting the map data.

[0149] The second determining module 520 is used to determine the first current scene category based on the current environmental information and vehicle status information of the target vehicle when the map data is unavailable.

[0150] The third determining module 530 is used to determine the target perception motion characteristics of the target object based on the self-vehicle status information and the target vehicle's object perception information.

[0151] The recognition module 540 is used to identify the intent of the target object based on a preset rule table, according to the first current scene category and the target perception motion characteristics.

[0152] In one possible implementation, the vehicle status information includes the vehicle's position coordinates and orientation. The second determining module 520 is specifically used to: divide the current driving environment of the target vehicle into regions based on the vehicle's position coordinates and current environment information to obtain the target region corresponding to the target vehicle; and determine the first current scene category based on the regional road features of the target region and the vehicle's orientation.

[0153] In one possible implementation, the object perception information includes at least one first object, as well as state information and category identification information of each first object. The third determining module 530 is specifically used to: filter at least one first object according to the vehicle state information and the state information and category identification information of each first object to obtain a target object; and calculate the target perception motion features of the target object according to the state information of the target object and the vehicle state information.

[0154] In one possible implementation, the recognition module 540 is specifically used to: generate structured data based on the first current scene category and the target perceived motion characteristics; match the structured data in a preset rule table, and output the intent type of the target object and the confidence level of the intent type.

[0155] In one possible implementation, the second determining module 520 is further configured to: determine a second current scene category based on the map data and the vehicle status information when the map data is available; identify the intent of the target object based on the second current scene category; or identify the intent of the target object based on the second current scene category and the target's perceived motion characteristics.

[0156] In one possible implementation, the second determining module 520 is further configured to: perform coordinate transformation on the latitude and longitude of the map data to obtain the global plane coordinates corresponding to the map data; perform coordinate transformation on the global plane coordinates to obtain the local plane coordinate system corresponding to the target vehicle; and determine the second current scene category based on the local plane coordinates and the vehicle status information.

[0157] In one possible implementation, the preset rule table is constructed as follows: A sample dataset is acquired, including road topology data, vehicle data, multiple different second objects, and the initial motion trajectory and perception data of each second object; for each of the multiple different second objects, ground truth annotation is performed on the initial motion trajectory of the second object to obtain the target motion trajectory of the second object, which carries the ground truth intent type of the second object; based on the road topology data, vehicle data, and perception data of the second objects, the perception motion features and topological features of the second objects are extracted; based on the ground truth intent type grouping, the feature combination rules of different ground truth intent types are identified according to the ground truth intent type, topological features, and perception motion features of each second object; and based on the feature combination rules, a preset rule table is constructed.

[0158] The intent recognition device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0159] This application also provides a vehicle, including a vehicle body and an intent recognition system, which is used to implement the method provided in the above-described method embodiments. The intent recognition system is as described above. Figure 1 As shown in the diagram, it will not be repeated here.

[0160] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the device 60 further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0161] In a specific implementation, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to perform the above-described method.

[0162] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0163] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0164] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device.

[0165] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0166] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0167] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0168] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0169] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0170] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0171] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0172] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0173] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0174] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0175] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An intent recognition method, characterized in that, include: In response to the target vehicle detecting map data, it is determined whether the map data is available; If the map data is unavailable, the first current scene category is determined based on the target vehicle's current environmental information and vehicle status information. Based on the vehicle status information and the target vehicle's object perception information, the target object's target perception motion characteristics are determined; Based on a preset rule table, the intent of the target object is identified according to the first current scene category and the target perceived motion characteristics.

2. The intent recognition method according to claim 1, characterized in that, The vehicle status information includes the vehicle's position coordinates and orientation. Determining the first current scene category based on the target vehicle's current environment information and the vehicle status information includes: Based on the vehicle's position coordinates and the current environment information, the current driving environment of the target vehicle is divided into regions to obtain the target region corresponding to the target vehicle; The first current scene category is determined based on the regional road characteristics of the target area and the vehicle's orientation.

3. The intent recognition method according to claim 1, characterized in that, The object perception information includes at least one first object, and state information and category identification information for each first object. The step of determining the target perception motion characteristics of the target object based on the vehicle's state information and the target vehicle's object perception information includes: Based on the vehicle status information and the status information and category identification information of each first object, the at least one first object is filtered to obtain the target object; Based on the state information of the target object and the state information of the vehicle, the target perception motion features of the target object are calculated.

4. The intent recognition method according to claim 1, characterized in that, The step of identifying the intent of the target object based on a preset rule table, according to the first current scene category and the target perceived motion features, includes: Based on the first current scene category and the target perceived motion features, structured data is generated; The structured data is matched against the preset rule table, and the intent type of the target object and the confidence level of the intent type are output.

5. The intent recognition method according to any one of claims 1 to 4, characterized in that, Also includes: If the map data is available, then a second current scene category is determined based on the map data and the vehicle status information; Identify the intent of the target object based on the second current scene category; Alternatively, the intent of the target object can be identified based on the second current scene category and the target perceived motion characteristics.

6. The intent recognition method according to claim 5, characterized in that, The step of determining the second current scene category based on the map data and the vehicle status information includes: The latitude and longitude of the map data are transformed to obtain the global planar coordinates corresponding to the map data; The global plane coordinates are transformed to obtain the local plane coordinate system corresponding to the target vehicle; The second current scene category is determined based on the local planar coordinates and the vehicle status information.

7. The intent recognition method according to any one of claims 1 to 4, characterized in that, The preset rule table is constructed in the following way: Obtain a sample dataset, which includes road topology data, vehicle data, multiple different second objects, and the initial motion trajectory and perception data of each second object; For each of the multiple different second objects, the initial motion trajectory of the second object is labeled with truth values ​​to obtain the target motion trajectory of the second object, and the target motion trajectory carries the truth intention type of the second object; Based on the road topology data, the vehicle data, and the perception data of the second object, extract the perception motion features and topology features of the second object; Based on truth intent type grouping, the feature combination rules of different truth intent types are identified according to the truth intent type of each second object, the topological features and the perceptual motion features; Based on the feature combination rules, the preset rule table is constructed.

8. A vehicle, characterized in that, include: The vehicle body and the intent recognition system, wherein the intent recognition system is used to implement the intent recognition method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the intent recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the intent recognition method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, include: A computer program that, when executed by a processor, implements the intent recognition method as described in any one of claims 1 to 7.

Images