Track prediction method and device of non-motor vehicle, medium and product

Abstract

The invention provides a non-motor vehicle trajectory prediction method and device, a computer readable storage medium and a computer program product. The method comprises the steps of determining a target non-motor vehicle under a right turn intention of a motor vehicle; obtaining current driving information, a current position and a target passing intention of a target non-motor vehicle; the current driving information comprises a current speed and a current course; determining a prediction track set of the target non-motor vehicle based on the current driving information and the current position through a target prediction model; determining a candidate track set based on the right turn reference track, the current driving information, the target passing intention and the current position of the motor vehicle; each candidate track in the candidate track set has no collision risk with a first object except the target non-motor vehicle in the environment of the motor vehicle; and determining a target trajectory from the candidate trajectory set based on the predicted trajectory set and the candidate trajectory set.

Description

Technical Field

[0001] This application relates to the field of automotive technology, specifically to a method, device, computer-readable storage medium, and computer program product for predicting the trajectory of non-motorized vehicles. Background Technology

[0002] In intelligent driving systems, trajectory prediction technology is a crucial component for enabling autonomous decision-making and safe path planning by motor vehicles. Especially in complex traffic scenarios, such as intersections, the unpredictable behavior of non-motorized vehicles poses a significant challenge to the predictive capabilities of motor vehicles. Accurately predicting the future trajectories of non-motorized vehicles helps improve the efficiency and ride comfort of driver assistance systems.

[0003] However, among related technologies, a typical method is to predict the trajectory of non-motorized vehicles based on driving rules. However, this prediction method is difficult to efficiently process a large number of features and is easily affected by fluctuations in upstream input, thus reducing the prediction accuracy of non-motorized vehicle trajectories. Another typical method is to predict the trajectory of non-motorized vehicles based on big data models. However, this prediction method is difficult to guarantee long-term prediction results, and the predicted trajectory may not meet kinematic or safety constraints, thus reducing the prediction accuracy of non-motorized vehicle trajectories. Summary of the Invention

[0004] This application provides a method, apparatus, computer-readable storage medium, and computer program product for predicting the trajectory of non-motorized vehicles, in order to solve the problem of low prediction accuracy of the trajectory of non-motorized vehicles in the prior art.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows: A method for predicting the trajectory of a non-motorized vehicle includes: identifying a target non-motorized vehicle with a right-turn intention; acquiring the current driving information, current position, and target traffic intention of the target non-motorized vehicle; the current driving information includes current speed and current heading; determining a predicted trajectory set for the target non-motorized vehicle based on the current driving information and current position using a target prediction model; determining a candidate trajectory set based on the right-turn reference trajectory of the motor vehicle, the current driving information, the target traffic intention, and the current position; ensuring that each candidate trajectory in the candidate trajectory set has no collision risk with a first object in the environment where the motor vehicle is located, excluding the target non-motorized vehicle; and determining the target trajectory from the candidate trajectory set based on the matching degree between the predicted trajectory set and the candidate trajectory set.

[0006] Based on the aforementioned technical methods, firstly, by identifying the target non-motorized vehicle, the resource waste of predicting and calculating irrelevant non-motorized vehicles can be reduced, thus saving unnecessary resource expenditure. Secondly, based on the target non-motorized vehicle's current driving information and current position, a predicted trajectory set is generated. This is further combined with a right-turn reference trajectory, environmental information, and the target's communication intent to generate a candidate trajectory set that meets kinematic constraints and safety requirements. Thus, by considering road environment characteristics and the target non-motorized vehicle's intended travel, a targeted candidate trajectory set can be generated, avoiding unnecessary computational overhead and improving the accuracy and safety of trajectory prediction. Finally, the candidate trajectory set is evaluated using the predicted trajectory set, and the target trajectory is selected from the candidate trajectory set. This allows the target trajectory to simultaneously combine the advantages of rule constraints and model prediction, improving the computational accuracy of the target trajectory and thereby enhancing the accuracy and robustness of trajectory prediction.

[0007] Furthermore, based on the right-turn reference trajectory of the motor vehicle, current driving information, target traffic intention, and current location, a candidate trajectory set is determined, including: determining a set of selected trajectories for the target non-motorized vehicle based on the right-turn reference trajectory, current driving information, target traffic intention, and current location; for each selected trajectory in the selected trajectory set, determining the expected acceleration of the target non-motorized vehicle in each sampling period within the sampling duration using a Gaussian decay method based on the current speed; for each selected trajectory in the selected trajectory set, determining the sampling points for each sampling period on the selected trajectory based on the expected acceleration and current speed in each sampling period; and for each selected trajectory in the selected trajectory set, determining a candidate trajectory in the candidate trajectory set based on multiple sampling points within the sampling duration and the location information of the first object.

[0008] Based on the aforementioned technical methods, the acceleration trend of the target non-motorized vehicle is simulated using the Gaussian decay method to generate sampling points that better reflect actual driving behavior. These points are then combined with the location information of the first object to eliminate trajectories with collision risks from the selected trajectory set, thereby improving the safety and rationality of the candidate trajectory set. Compared to traditional methods that assume fixed acceleration or are purely data-driven, this approach better fits the dynamic behavior of the target non-motorized vehicle and reduces the generation of unreasonable trajectories.

[0009] Furthermore, the current heading includes the current heading angle; the target's traffic intention includes a crossing intention and a right turn intention; based on the right turn reference trajectory, current driving information, target traffic intention, and current position, a set of selected trajectories for the target non-motorized vehicle is determined, including: for a crossing intention, determining the first direct trajectory corresponding to the current heading angle; for a right turn intention, determining at least one second trajectory for the target non-motorized vehicle based on the right turn reference trajectory, current heading, current speed, and current position; and based on at least one second trajectory and the first direct trajectory, a set of selected trajectories is determined. Based on the aforementioned technical means, when the target's travel intention includes both crossing intention and right-turn intention, different trajectories are constructed according to the crossing intention and right-turn intention respectively. This method can more comprehensively cover the travel trajectory of the target non-motorized vehicle, improve the accuracy of predicting the travel trajectory of the target non-motorized vehicle, and thus enhance the decision-making and planning ability of motor vehicles in right-turn scenarios.

[0010] Furthermore, based on the right-turn reference trajectory, current heading, current speed, and current position, at least one second trajectory for the target non-motorized vehicle is determined, including: determining first control point information based on the current position and the right-turn reference trajectory; the first control point information includes a first distance and a second distance; the first distance represents the shortest Euclidean distance between the current position and the right-turn reference trajectory; the second distance represents the distance between the target point corresponding to the shortest Euclidean distance on the right-turn reference trajectory and the starting point of the right-turn reference trajectory; determining at least one second control point based on the first control point information, current heading, and current speed; determining third control point information corresponding to at least one second control point based on at least one second control point and the current speed; and determining at least one second trajectory based on at least one second control point, the third control point information corresponding to at least one second control point, the first control point information, and the current position.

[0011] Based on the aforementioned technical means, multi-level control points are generated based on the current position, right-turn reference trajectory, current heading, and current speed. At least one second trajectory is constructed through Bézier curve fitting. This method can more accurately simulate the right-turn behavior of the target non-motorized vehicle, effectively improve the accuracy and adaptability of the second trajectory prediction, thereby providing a reliable basis for the decision-making and planning of motor vehicles and improving the overall system's safety and traffic efficiency.

[0012] Furthermore, the method also includes: determining the lane type of the target non-motorized vehicle after it turns right and enters the lane; determining at least one second control point based on the first control point information, the current heading, and the current speed, including: determining a first duration based on the current heading and the target heading angle after the target non-motorized vehicle turns right; determining a third distance based on the first duration, a second distance, and the current speed; if the lane type is a non-motorized vehicle lane, determining multiple fourth distances based on the first distance and the width of the target non-motorized vehicle; determining multiple second control point information based on the third distance and the multiple fourth distances; if the lane type is a motorized vehicle lane, determining a fifth distance based on the first distance, the current speed, and the first duration; and determining a single second control point based on the third distance and the fifth distance.

[0013] Based on the aforementioned technical means, by distinguishing the lane type of the target non-motorized vehicle after turning right and entering the lane, and using different processing logics for the second control information, a single second control point is determined for the motorized vehicle lane scenario, while multiple second control point information is determined for the non-motorized vehicle lane scenario. This approach takes into account driving rules, generates a second trajectory that conforms to traffic rules more accurately, reduces the generation of invalid second trajectories, and improves the calculation efficiency and accuracy of the second trajectory.

[0014] Furthermore, determining the lane type of the target non-motorized vehicle after it turns right includes: determining the first lateral distance between the target object and the right-turn reference trajectory; the target object being located between the motorized vehicle lane and the non-motorized vehicle lane after the right turn; and determining the lane type based on the first lateral distance and the first distance.

[0015] Based on the aforementioned technical means, the type of lane the target non-motorized vehicle enters after turning right can be identified based on the first lateral distance and the first distance, thereby providing reasonable prior knowledge for the generation of the second trajectory and reducing the probability of generating a second trajectory that does not conform to traffic rules.

[0016] Furthermore, obtaining the target non-motorized vehicle's intended travel includes: when the target intersection between the first direct trajectory and the right-turn reference trajectory is located in a first area, determining that the target intended travel includes a right-turn intention; the first area refers to the area located on the side away from the motor vehicle from the first stop line; the first stop line is the stop line of the motor vehicle lane that the motor vehicle enters after turning right; when the target intersection is located in a second area, determining that the target intended travel includes a right-turn intention and a crossing intention; the second area is located on the side of the first stop line closer to the motor vehicle.

[0017] Based on the aforementioned technical means, the target travel intention of the non-motorized vehicle is determined by the geometric relationship between the first direct trajectory and the right-turn reference trajectory. This allows for accurate identification of whether the non-motorized vehicle intends to turn right or also cross the road, providing a basis for subsequent trajectory generation and thus improving the rationality and practicality of the prediction.

[0018] Furthermore, determining the target non-motorized vehicle when the motor vehicle intends to turn right includes: determining the current position of at least one first non-motorized vehicle; for each first non-motorized vehicle, determining whether the first non-motorized vehicle is the target non-motorized vehicle based on its current position, the target coordinate system, and the right-turn reference trajectory; the target coordinate system has the first intersection point as the origin and the straight-ahead direction of the motor vehicle as the x-axis; the first intersection point is the intersection of the right-turn reference trajectory and the stop line of the lane where the motor vehicle is currently located.

[0019] Based on the aforementioned technical means, by establishing a unified target coordinate system and combining it with a right-turn reference trajectory, target non-motorized vehicles within the right-turn influence range of motor vehicles can be efficiently identified, providing an input basis for subsequent trajectory prediction.

[0020] Furthermore, based on the current position of the first non-motorized vehicle, the target coordinate system, and the right-turn reference trajectory, determining whether the first non-motorized vehicle is the target non-motorized vehicle includes: performing a vector transformation on the first coordinate of the target point in the target coordinate system to obtain a first vector; transforming the current position of the first non-motorized vehicle to the second coordinate in the target coordinate system to determine a second vector; determining whether the first non-motorized vehicle is located to the right of the right-turn reference trajectory based on the first and second vectors; if the first non-motorized vehicle is located to the right of the right-turn reference trajectory, transforming the current heading angle of the first non-motorized vehicle to the target coordinate system to obtain a target heading angle; and determining whether the first non-motorized vehicle is the target non-motorized vehicle based on whether the target heading angle meets a preset angle threshold range.

[0021] Based on the aforementioned technical methods, firstly, by vectorizing the target point, it is easier to compare it with the position vector of the first non-motorized vehicle, thereby determining whether the first non-motorized vehicle is located to the right of the right-turn reference trajectory. Then, if the first non-motorized vehicle is located to the right of the right-turn reference trajectory, its current heading angle is used to further determine whether it is the target non-motorized vehicle. This dual verification method improves the accuracy of target non-motorized vehicle identification.

[0022] A trajectory prediction device for non-motorized vehicles includes: a first determining module for determining a target non-motorized vehicle with a right-turn intention of a motor vehicle; acquiring the current driving information, current position, and target traffic intention of the target non-motorized vehicle; the current driving information includes current speed and current heading; a second determining module for determining a predicted trajectory set of the target non-motorized vehicle based on the current driving information and current position using a target prediction model; a third determining module for determining a candidate trajectory set based on the right-turn reference trajectory of the motor vehicle, the current driving information, the target traffic intention, and the current position; each candidate trajectory in the candidate trajectory set has no collision risk with a first object in the environment where the motor vehicle is located, excluding the target non-motorized vehicle; and a fourth determining module for determining the target trajectory from the candidate trajectory set based on the matching degree between the predicted trajectory set and the candidate trajectory set.

[0023] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0024] A computer program product includes a computer program or instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0025] The beneficial effects of this application are: (1) First, by identifying the target non-motorized vehicle, the resource waste of predicting and calculating irrelevant non-motorized vehicles can be reduced, thus saving unnecessary resource expenditure. Then, based on the current driving information and current position of the target non-motorized vehicle, a predicted trajectory set is generated. Furthermore, by combining the right-turn reference trajectory, environmental information, and the target's communication intention, a candidate trajectory set that meets kinematic constraints and safety requirements is generated. In this way, by considering the road environment characteristics and the target non-motorized vehicle's target travel intention, a candidate trajectory set can be generated in a targeted manner, avoiding unnecessary computational overhead and improving the accuracy and safety of trajectory prediction. Finally, the candidate trajectory set is evaluated by the predicted trajectory set, and the target trajectory is selected from the candidate trajectory set. This allows the target trajectory to combine the advantages of rule constraints and model prediction, improving the computational accuracy of the target trajectory and thus improving the accuracy and robustness of trajectory prediction.

[0026] (2) Different trajectories are constructed according to the intention to cross and the intention to turn right. This method can more comprehensively cover the driving trajectory of the target non-motorized vehicle, improve the accuracy of predicting the driving trajectory of the target non-motorized vehicle, and thus enhance the decision-making and planning ability of motor vehicles in the right-turn scenario. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the implementation process of the trajectory prediction method for non-motorized vehicles proposed in this application. Figure 1 ; Figure 2 This is a schematic diagram of the trajectory prediction result for a non-motorized vehicle proposed in this application; Figure 3 This is a schematic diagram of the implementation process of the trajectory prediction method for non-motorized vehicles proposed in this application. Figure 2 ; Figure 4 This is a schematic diagram of the composition structure of a trajectory prediction device for non-motorized vehicles proposed in this application.

[0028] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] This application provides a method for predicting the trajectory of non-motorized vehicles, such as... Figure 1 As shown, the trajectory prediction method includes steps S100 to S130: Step S100: Determine the target non-motorized vehicle with the intention of turning right; obtain the current driving information, current location and target traffic intention of the target non-motorized vehicle; the current driving information includes the current speed and current heading.

[0031] In this application, before performing this step, it is also necessary to determine whether the vehicle's intention to travel is to turn right. The determination that the vehicle intends to turn right can be achieved in any of the following ways: Method 1, the vehicle is currently in the right turn waiting area at the intersection; Method 2, the vehicle has turned on its right turn signal; Method 3, the navigation system on the vehicle shows a right turn instruction; Method 4, the vehicle is traveling on a road with a right turn instruction.

[0032] The target non-motorized vehicle can refer to a non-motorized vehicle that is traveling on the right side of a motorized vehicle and may interact with the motorized vehicle when the motorized vehicle intends to turn right.

[0033] Current driving information can describe the motion state of the target non-motorized vehicle at the current moment. This motion state can be determined by the current speed and current heading. The current speed can include the current travel speed and current travel acceleration, and the current heading can include the current heading angle and the rate of change of the current heading angle. In some embodiments, a sensing module in the vehicle perceives the environment in which the vehicle is currently located, and then determines the current driving information of the target non-motorized vehicle based on the perceived information.

[0034] The target communication intention refers to the direction that the target non-motorized vehicle may take in the current traffic environment, such as turning right along with the vehicle. In some implementations, the target communication intention can be predicted by the vehicle based on perceived information, or the perceived information may contain the target communication intention.

[0035] In some implementations, the target traffic intention may include a right-turn intention and a cross-traffic intention, or a right-turn intention.

[0036] Step S110: Determine the predicted trajectory set of the target non-motorized vehicle based on the current driving information and current location using the target prediction model.

[0037] Here, the target prediction model is a pre-trained machine learning model. In this application, the target prediction model is trained based on historical data and has the ability to learn the behavior patterns of non-motorized vehicles. Thus, after inputting the current driving information, current position, and the current driving information and current position of the motor vehicle into the target prediction model, the target prediction model can predict the possible driving trajectory of the target non-motorized vehicle in the future period of time, as well as a probability value corresponding to each driving trajectory. The probability value represents the likelihood of the driving trajectory occurring.

[0038] The predicted trajectory set is a set of multimodal predicted trajectories output by the target prediction model, with each predicted trajectory representing a possible future driving path.

[0039] Step S120: Based on the right-turn reference trajectory of the motor vehicle, current driving information, target communication intent and current location, determine a set of candidate trajectories; each candidate trajectory in the set of candidate trajectories has no collision risk with the first object in the environment where the motor vehicle is located, excluding the target non-motor vehicle.

[0040] Here, the right-turn reference trajectory refers to the preset path or navigation reference line followed by a motor vehicle during a right turn. It is usually provided by the navigation system or map data. It can be understood that the right-turn reference trajectory reflects the expected driving route of the motor vehicle during the right turn and is an important basis for generating candidate trajectories.

[0041] The candidate trajectory set is a set of candidate trajectories that conform to the logic of the traffic scenario, generated based on rule constraints (such as kinematic rationality, no collision risk, etc.). Each candidate trajectory in the candidate trajectory set must meet the requirement of not conflicting with other objects in the environment (such as static obstacles, boundaries, etc.).

[0042] Step S130: Determine the target trajectory from the candidate trajectory set based on the matching degree between the predicted trajectory set and the candidate trajectory set.

[0043] The target trajectory is the trajectory in the candidate trajectory set that is most likely to reflect the actual behavior of the target non-motorized vehicle.

[0044] In some implementations, the matching degree between each candidate trajectory in the candidate trajectory set and the predicted trajectory set is determined; based on the matching degrees corresponding to all candidate trajectories, the target trajectory is determined. The matching degree between each candidate trajectory and the predicted trajectory can be determined based on the similarity between each candidate trajectory and the predicted trajectories in the predicted trajectory set.

[0045] In some implementations, the probability information of each predicted trajectory in the predicted trajectory set can also be determined through the target model. In this scenario, firstly, the similarity between each candidate trajectory and each predicted trajectory in the predicted trajectory set is determined; then, all similarities are multiplied by their corresponding probability information and summed to obtain a weighted similarity; based on the weighted similarities corresponding to all candidate trajectories, the target trajectory is determined. The similarity between the candidate trajectory and the predicted trajectory can be achieved in the following ways: Method 1: Calculate the Euclidean distance between the candidate trajectory and the predicted trajectory; Method 2: Calculate the Hausdorff distance between the candidate trajectory and the predicted trajectory; Method 3: Calculate the average target distance error between the candidate trajectory and the predicted trajectory.

[0046] The following describes the method for determining the target trajectory, taking the average distance error of the target as an example. It may include steps S131 and S132: Step S131: For each candidate trajectory in the candidate trajectory set, based on the trajectory distance calculation results between each predicted trajectory in the predicted trajectory set and the candidate trajectory, determine the target average distance error of the candidate trajectory. Here, the predicted trajectory set is a collection of multimodal trajectories generated by a pre-trained target prediction model. Each predicted trajectory corresponds to a possible driving path for the target non-motorized vehicle. The candidate trajectory set consists of several reasonable trajectory paths determined based on environmental information (such as road boundaries and static obstacles) and driving experience rules.

[0047] Trajectory distance calculation refers to quantifying the spatial difference between each predicted trajectory and the candidate trajectory, usually using the average distance error (ADE) over time series as a metric.

[0048] In some implementations, for each identical time point, a first trajectory point of the predicted trajectory at the first time point and a second trajectory point of the candidate trajectory at the first time point are determined; a first calculated distance between the first trajectory point and the second trajectory point is determined; based on the first calculated distances corresponding to multiple time points within a preset calculation time period, the average distance error between the predicted trajectory and the candidate trajectory is determined, and this average distance error is the trajectory distance calculation result mentioned above. The preset calculation time period can be 4 seconds.

[0049] It is understood that this application requires calculating the trajectory distance between each candidate trajectory and all predicted trajectories in the predicted trajectory set, thus obtaining multiple average distance errors. Based on these multiple average distance errors, the target average distance error can be obtained. The method for determining the target average distance error may include the following: Method 1: If the target prediction model can also output the probability weight of each predicted trajectory, multiply each average distance error by its corresponding probability weight to obtain a first value; sum all the first values ​​to obtain the target average distance error. The probability weight is used to represent the confidence level of the predicted trajectory.

[0050] Method 2: When the target prediction model does not output the probability weight of each predicted trajectory, sum all the average distance errors to obtain the target average distance error.

[0051] Method 3: When the target prediction model does not output the probability weight of each predicted trajectory, sum all the average distance errors to obtain a second value; divide the second value by the number of all average distance errors to obtain the target average distance error.

[0052] In this application, the matching degree between the candidate trajectory and the predicted trajectory set can be evaluated by the target average distance error. The smaller the target average distance error, the higher the matching degree between the candidate trajectory and the model output trajectory.

[0053] Step S132: Based on the minimum value among all target average distance errors, determine the target trajectory from the candidate trajectory set.

[0054] In this application, the target trajectory corresponds to the minimum of all target average distance errors, which is used to measure the trajectory similarity between the candidate trajectory and the predicted trajectory set. The smaller the target average distance error, the higher the trajectory similarity between the candidate trajectory and the predicted trajectory set. This ensures that the target trajectory not only meets the environmental constraints but also has a high similarity to the prediction results of the target prediction model, thus possessing higher reliability.

[0055] In this embodiment, by calculating the trajectory distance between the candidate trajectories generated by the rules and the predicted trajectory set predicted by the model, the average target distance error between each candidate trajectory and the predicted trajectory set can be determined. The trajectory with the smallest average target distance error is then selected as the final target trajectory from all candidate trajectories. This method achieves an effective balance between rule constraints and model prediction advantages, thereby improving the accuracy and robustness of target trajectory prediction.

[0056] Understandably, the target trajectory can serve as a crucial input for intelligent driving in motor vehicles, determining their obstacle avoidance strategies and path adjustments during right turns. For instance, if the target trajectory indicates that the non-motorized vehicle will continue straight across the intersection, the vehicle needs to slow down or adjust its right-turn path in advance to avoid a collision. If the predicted data indicates that the non-motorized vehicle will turn right to follow the vehicle, the vehicle can accelerate its right turn while ensuring safety, thereby improving traffic efficiency.

[0057] Therefore, the accuracy of target trajectory prediction plays a decisive role in the overall computational accuracy of the intelligent driving system.

[0058] In summary, the non-motorized vehicle trajectory prediction method provided in this application achieves efficient, reasonable, and safe trajectory prediction by combining the advantages of rule constraints and model prediction. By incorporating road environment features and modeling the travel intentions of target non-motorized vehicles, the system can generate different types of candidate trajectories in a targeted manner, thereby avoiding unnecessary computational overhead and improving the accuracy and safety of trajectory prediction. The system also uses a prediction model to score candidate trajectories and selects the optimal trajectory as the final output, further enhancing the stability and reliability of trajectory prediction.

[0059] In this embodiment, firstly, by identifying the target non-motorized vehicle, the resource waste of predicting and calculating irrelevant non-motorized vehicles can be reduced, thus saving unnecessary resource overhead. Then, based on the current driving information and current position of the target non-motorized vehicle, a predicted trajectory set is generated. Furthermore, by combining a right-turn reference trajectory, environmental information, and the target's communication intent, a candidate trajectory set that meets kinematic constraints and safety requirements is generated. This approach, by considering road environment characteristics and the target non-motorized vehicle's intended travel, allows for the targeted generation of candidate trajectory sets, avoiding unnecessary computational overhead and improving the accuracy and safety of trajectory prediction. Finally, the candidate trajectory set is evaluated using the predicted trajectory set, and the target trajectory is selected from the candidate trajectory set. This approach combines the advantages of rule constraints and model prediction, improving the computational accuracy of the target trajectory and thus enhancing the accuracy and robustness of trajectory prediction.

[0060] In some embodiments, step S120 may include steps S121 to S124: Step S121: Based on the right turn reference trajectory, current driving information, target traffic intention and current location, determine the set of filtered trajectories for the target non-motorized vehicle; In some implementations, the set of filtered trajectories corresponds to the target communication intent. For example, if the target communication intent is a right turn intent, the set of filtered trajectories includes at least one filtered trajectory corresponding to the right turn intent; or, if the target communication intent is a cross-traffic intent and a right turn intent, the set of filtered trajectories includes at least one filtered trajectory corresponding to the right turn intent and one filtered trajectory corresponding to the cross-traffic intent.

[0061] Understandably, the filtered trajectory set reflects the potential movement trend of the target non-motorized vehicle under the current road characteristics and target communication intent. However, the filtered trajectory set does not consider the relationship between the target non-motorized vehicle and the first object in the current environment when it is driving. Therefore, it is necessary to further filter the filtered trajectory set to remove trajectories in the filtered trajectory set that have a collision risk with the first object, thereby improving the safety of the obtained candidate trajectory set.

[0062] Step S122: For each selected trajectory in the selected trajectory set, based on the current speed, determine the expected acceleration of the target non-motorized vehicle in each sampling period within the sampling time using the Gaussian decay method; Here, the Gaussian attenuation method is an attenuation strategy based on the Gaussian function (normal distribution function). Its core is to construct a smooth bell-shaped curve to achieve the effect that the correlation, weight or signal strength and other indicators gradually decrease as the distance from the ideal point (or reference point) increases.

[0063] In some implementations, the current speed includes the current acceleration. Thus, by using a Gaussian decay method to extrapolate the current acceleration, the expected acceleration of the target non-motorized vehicle in each sampling period can be obtained; the expected acceleration differs in different sampling periods. It is understood that the acceleration variation of the target non-motorized vehicle in different sampling periods can be simulated using the Gaussian decay method.

[0064] In practice, using the current acceleration as the basic parameter, a Gaussian function is used to extrapolate future acceleration, causing the acceleration values ​​within different sampling periods to gradually decrease over time. It is understandable that as the target non-motorized vehicle approaches the intersection, it will often decelerate to observe traffic conditions. This application introduces a Gaussian attenuation method to more closely approximate the driving behavior of the target non-motorized vehicle as it approaches the intersection, thus improving the accuracy of the candidate trajectory set.

[0065] In some implementations, the sampling period and sampling duration can be preset, such as setting the sampling duration to 8s or 0.1s.

[0066] Step S123: For each selected trajectory in the selected trajectory set, based on the expected acceleration and current velocity of each sampling period, determine the sampling points on the selected trajectory for each sampling period; In some implementations, the current speed includes the current travel speed. In the first sampling period, based on the current travel speed, the time corresponding to the first sampling period, and the expected acceleration corresponding to the first sampling period, a first travel distance of the target non-motorized vehicle within the first sampling period can be obtained, thereby determining a first sampling point on the screening trajectory that matches the first travel distance. In the second sampling period, based on the current travel speed corresponding to the first sampling point, the time corresponding to the second sampling period, and the expected acceleration corresponding to the second sampling period, a second travel distance of the target non-motorized vehicle within the second sampling period can be obtained, thereby determining a second sampling point on the screening trajectory that matches the second travel distance. This process is repeated to obtain multiple sampling points within the sampling time. The current travel speed corresponding to the first sampling point can be determined based on the current travel speed, the time corresponding to the first sampling period, and the expected acceleration corresponding to the first sampling period.

[0067] In some implementations, when the sampling duration is set to 8 seconds or 0.1 seconds, 80 sampling points can be obtained.

[0068] In this application, a trajectory description of the target non-motorized vehicle can be formed by generating multiple sampling points on each screening trajectory.

[0069] Step S124: For each selected trajectory in the selected trajectory set, based on multiple sampling points within the sampling time and the location information of the first object, determine the candidate trajectory in the candidate trajectory set.

[0070] Here, the first object can refer to static obstacles and other traffic participants in the current environment of the motor vehicle.

[0071] In some implementations, for each sampling point, the location information of the sampling point is matched with the location information of the first object; if the match is successful, the candidate trajectory is removed from the set of selected trajectories.

[0072] In some implementations, if the location information corresponding to multiple sampling points does not match the location information of the first object, the filtered trajectory is determined as the candidate trajectory.

[0073] In some implementations, multiple sampling points within a target sampling duration are acquired; based on the multiple sampling points within the target duration and the position information of the first object, candidate trajectories in the candidate trajectory set are determined. The target sampling duration is shorter than the sampling duration; for example, the target sampling duration can be set to 4 seconds.

[0074] In this application, the relative positional relationship between the trajectory formed by the sampling point and the surrounding static obstacles (such as green belts and curbs) and other traffic participants is comprehensively considered. Trajectories with no collision risk are selected as the final candidate trajectory set. This method can eliminate unreasonable trajectories that may collide in a short period of time, thereby ensuring that the output trajectory is safe and feasible.

[0075] In this embodiment, the acceleration variation trend of the target non-motorized vehicle is simulated using the Gaussian decay method to generate sampling points that better reflect actual driving behavior. These points are then combined with the location information of the first object to eliminate trajectories with collision risks from the selected trajectory set, thereby improving the safety and rationality of the candidate trajectory set. Compared to traditional methods that assume fixed acceleration or are purely data-driven, this approach better fits the dynamic behavior of the target non-motorized vehicle and reduces the generation of unreasonable trajectories.

[0076] In some embodiments, the current heading includes the current heading angle; the target passage intention includes a crossing intention and a right turn intention; The above step S121 may include steps S1211 and S1213: Step S1211: Based on the intention to cross, determine the first direct trajectory corresponding to the current heading angle; Here, the current heading angle describes the angle between the current direction of movement of the target non-motorized vehicle and the reference direction. Crossing intent is used to characterize the target non-motorized vehicle's intention to cross the intersection, such as going straight or turning left.

[0077] The first direct-fire trajectory is a straight-line trajectory directly derived from the target non-motorized vehicle's current heading angle. In other words, the first direct-fire trajectory is generated based on the assumption that the target non-motorized vehicle will continue to move at a constant speed in its current direction. Figure 2 As shown, Figure 2 The image shows four target non-motorized vehicles (vru1, vru2, vru3, and vru4). Among them, the P21 trajectory line derived from vru2 and the P31 trajectory line derived from vru3 are the first direct trajectory.

[0078] Step S1212: Based on the right turn intention, determine at least one second trajectory for the target non-motorized vehicle, according to the right turn reference trajectory, current heading, current speed and current position; The second trajectory is used to describe the possible right-turn trajectory of the target non-motorized vehicle when it turns right. It should be noted that since the target non-motorized vehicle may have multiple reasonable right-turn trajectories, at least one second trajectory can be determined in this scenario.

[0079] The trajectory selection process involves choosing the optimal trajectory from all candidate trajectories, taking into account both the first and second direct-fire trajectories. This selection process ensures the system can accommodate reasonable trajectories for different travel intentions and avoids overlooking critical situations.

[0080] Step S1213: Determine the set of selected trajectories based on at least one second trajectory and the first direct trajectory.

[0081] In some implementations, the set of at least one second trajectory and the first direct trajectory is determined as the filter trajectory set.

[0082] In this embodiment of the application, when the target traffic intention includes both crossing intention and right turn intention, different trajectories are constructed according to the crossing intention and right turn intention respectively. Through this method, the driving trajectory of the target non-motorized vehicle can be more comprehensively covered, the accuracy of the predicted driving trajectory of the target non-motorized vehicle can be improved, thereby enhancing the decision-making and planning ability of motor vehicles in right turn scenarios.

[0083] In some embodiments, the current heading includes the current heading angle; the target passage intention includes a right turn intention; Step S121 above may include steps S1214 and S1215: Step S1214: Based on the right-turn reference trajectory, current heading, current speed, and current position, determine at least one second trajectory for the target non-motorized vehicle; Step S1215: Determine the set of selected trajectories based on at least one second trajectory and the first direct trajectory.

[0084] In some implementations, the set of at least one second trajectory and the first direct trajectory is determined as the filter trajectory set.

[0085] In this embodiment of the application, when the target traffic intention includes a right turn intention, a corresponding trajectory is constructed based on the right turn intention. This method can more comprehensively cover the driving trajectory of the target non-motorized vehicle, improve the accuracy of the predicted driving trajectory of the target non-motorized vehicle, and thus enhance the decision-making and planning capabilities of motor vehicles in right turn scenarios.

[0086] In some embodiments, step S1212 may include steps S200 to S230: Step S200: Based on the current position and the right-turn reference trajectory, determine the first control point information; the first control point information includes a first distance and a second distance; the first distance is used to represent the shortest Euclidean distance between the current position and the right-turn reference trajectory; the second distance is used to represent the distance between the target point corresponding to the shortest Euclidean distance on the right-turn reference trajectory and the starting point of the right-turn reference trajectory; Here, the first control point information is used to describe the spatial relationship between the target non-motorized vehicle and the right-turn reference trajectory, including two key parameters: the first distance ( ) and second distance ( ).in, The first distance represents the shortest Euclidean distance from the current position of the target non-motorized vehicle to the right-turn reference trajectory, reflecting the degree to which the current position of the target non-motorized vehicle deviates from the right-turn reference trajectory.

[0087] The second distance represents the distance between the point on the right-turn reference trajectory that has the shortest Euclidean distance to the current position of the target non-motorized vehicle and the starting point of the right-turn reference trajectory. It is used to quantify the relative position of the current position of the target non-motorized vehicle on the right-turn reference trajectory.

[0088] Step S210: Based on the first control point information, the current heading, and the current speed, determine at least one second control point; Here, the second control point information is used to describe the driving position of the target non-motorized vehicle when it enters the lane in a straight manner after completing a right turn. The second control point information is the control point information derived from the first control point information, the current heading, and the current speed.

[0089] In some implementations, the second control point information may include two key parameters: a third distance ( ), fourth distance or fifth distance ( Thus, the information of the second control point can be characterized as ,in, It represents the shortest Euclidean distance from the position corresponding to the second control point information to the right-turn reference trajectory, reflecting the degree to which the position corresponding to the second control point information deviates from the right-turn reference trajectory.

[0090] This represents the distance between the point on the right-turn reference trajectory that has the shortest Euclidean distance to the starting point of the right-turn reference trajectory, and is used to quantify the relative position of the position corresponding to the second control point information on the right-turn reference trajectory.

[0091] Understandably, the second trajectory is used to describe the right-turn trajectory of the target non-motorized vehicle when it turns right. The right-turn trajectory includes the first sub-trajectory before the target non-motorized vehicle's direction of travel is parallel to the lane after the right turn, and the second sub-trajectory after the first parallelism. The second control point information represents the intersection point of the first sub-trajectory and the second sub-trajectory.

[0092] Step S220: Based on at least one second control point information and the current speed, determine the third control point information corresponding to each of the at least one second control point information; Here, the third control point information is used to describe the driving position of the target non-motorized vehicle entering the lane in a straight manner. It can be understood that the second trajectory is used to describe the right-turn driving trajectory of the target non-motorized vehicle when turning right, and the right-turn driving trajectory includes the first sub-trajectory and the second sub-trajectory. The third control point information is a point on the second sub-trajectory.

[0093] In some implementations, the third control point information can be characterized as ,in, This represents the distance between the point on the right-turn reference trajectory that has the shortest Euclidean distance to the position corresponding to the third control point information and the starting point of the right-turn reference trajectory. It is used to quantify the relative position of the position corresponding to the third control point information on the right-turn reference trajectory.

[0094] In some implementations, based on the current driving speed, preset duration, and... ,Sure See the following formula (1): (1); in, Indicates the preset duration. You can choose 0.1s. This represents the current driving speed.

[0095] Understandably, the third control point information is derived from the second control point information using the current speed. The third control point information is used to refine the trajectory shape. The current speed is used as an input parameter to determine the motion state of the target non-motorized vehicle at different points in time. Changes in this motion state will further affect the curvature and extension direction of the trajectory.

[0096] Step S230: Based on at least one second control point information, the third control point information corresponding to at least one second control point information, and the current position, determine at least one second trajectory.

[0097] In some implementations, each second trajectory is obtained by fitting a Bézier curve based on second control point information, third control point information, and the current position. Taking each second control point as an example, the first coordinate information of the second control point information in the target coordinate system and the second coordinate information of the corresponding third control point information in the target coordinate system are determined; the second trajectory is determined by fitting a Bézier curve based on the first coordinate information, the second coordinate information, and the current position.

[0098] In this embodiment, multi-level control points are generated based on the current position, right-turn reference trajectory, current heading, and current speed, and at least one second trajectory is constructed by Bézier curve fitting. Through the above method, the right-turn behavior of the target non-motorized vehicle can be simulated more accurately, which can effectively improve the accuracy and adaptability of the second trajectory prediction. This can provide a reliable basis for the decision-making and planning of motor vehicles and improve the safety and traffic efficiency of the overall system.

[0099] In some embodiments, the method further includes: Step S240: Determine the lane type that the target non-motorized vehicle will enter after turning right; Here, lane type refers to the nature of the road area that the target non-motorized vehicle enters after completing the right turn. Specifically, it can be divided into non-motorized vehicle lane type and motorized vehicle lane type. Non-motorized vehicle lane type refers to a dedicated road that only allows non-motorized vehicles to pass, and it is usually separated from the motorized vehicle lane by a median strip, while motorized vehicle lane type is a road that allows motorized vehicles to pass.

[0100] In some implementations, the lane type that the target non-motorized vehicle will enter after turning right can be determined based on the target non-motorized vehicle's current position and right-turn reference trajectory.

[0101] In some implementations, the number of predicted second trajectories varies depending on the lane type. For example, when the lane type is a motor vehicle lane, the number of second trajectories is set to 1, while when the lane type is a non-motor vehicle lane, the number of second trajectories can be greater than 1. The reason for setting fewer second trajectories for scenarios where the target non-motor vehicle turns right and enters the motor vehicle lane is twofold: firstly, it poses a significant safety hazard to the target non-motor vehicle driver, as the target non-motor vehicle generally does not travel in the motor vehicle lane, meaning the probability of this scenario being triggered is low; secondly, even if the target non-motor vehicle turns right into the motor vehicle lane, to reduce the probability of accidents, it will likely travel along the side of the motor vehicle lane, so trajectory prediction only needs to be performed according to this scenario.

[0102] In this application, a reasonable number of second trajectories can be generated specifically for different lane types, which can effectively avoid unnecessary waste of computing resources.

[0103] The above step S210 may also include steps S211 to S214: Step S211: Determine the first duration based on the current heading and the target heading angle after the target non-motorized vehicle turns right; In some implementations, the current heading may include the current heading angle and the rate of change of the current heading angle, wherein the rate of change of the current heading angle is a physical quantity describing how quickly the current heading angle changes over time.

[0104] The heading angle information of the target non-motorized vehicle after making a right turn refers to the heading angle information of the target non-motorized vehicle when it enters the right turn lane after completing the right turn. This heading angle information can be predicted based on the target non-motorized vehicle's current heading, current position, and surrounding environmental information collected by the vehicle; alternatively, it can be calculated based on lane information on a high-precision map.

[0105] The first duration is calculated by taking into account the difference between the target non-motorized vehicle's current direction of travel and its heading angle after turning right, and then combining this with the target non-motorized vehicle's turning rate to estimate the time required for the target non-motorized vehicle to complete the first sub-track.

[0106] In some implementations, a first difference is obtained by calculating the difference between the current heading angle and the target heading angle in the current heading; the quotient obtained by dividing the first difference by the rate of change of the current heading angle is determined as the first duration. For example, if the current heading angle is 90°, the target heading angle is 0°, and the target non-motorized vehicle rotates at an angle of 5° per second (i.e., the rate of change of the current heading angle), then the first duration is approximately 18 seconds.

[0107] In some implementations, a first difference is obtained by performing a difference calculation between the current heading angle and the target heading angle in the current heading; a first sub-duration is determined based on the first difference and the rate of change of the current heading angle; the smaller of the second preset duration and the first sub-duration is determined as the first duration, as shown in the following formula (2): = (2); in, Indicates the second preset duration. The value can be 2.0s.

[0108] It should be noted that determining the first duration is one of the key parameters in the prediction process of the second trajectory. The first duration directly affects the temporal and spatial distribution of subsequent trajectory sampling. By accurately estimating the first duration, the dynamic behavior of the target non-motorized vehicle during the turning process can be simulated more precisely, thereby improving the accuracy of the second trajectory prediction.

[0109] Step S212: Determine the third distance based on the first duration, the second distance, and the current speed; In some implementations, the third distance is determined using the following formula (3): (3); in, This represents the tangential speed obtained by projecting the current driving speed onto the right-turn reference trajectory.

[0110] In some implementations, the current driving speed is projected onto the right-turn reference trajectory to obtain... During implementation, the angle between the tangential direction at the target point and the current heading of the target non-motorized vehicle is determined. Based on this angle and the current speed, the current speed is decomposed into tangential velocity using the Pythagorean theorem. and normal velocity .

[0111] Step S213: When the lane type is a non-motorized vehicle lane, determine multiple fourth distances based on the first distance and the width of the target non-motorized vehicle; determine multiple second control point information based on the third distance and the multiple fourth distances; In some implementations, the fourth distance is determined as shown in the following formula (4): (4); in, The width of the target non-motorized vehicle is represented by , and n represents the sampling parameter, which takes the value of an integer.

[0112] It is understandable that multiple fourth distances can be obtained by setting n to different integer values, for example, hour, ; hour, .

[0113] In some implementations... It should also meet the target range, see the following formula (5): (5); in, This represents the shortest Euclidean distance between the left boundary of the non-motorized vehicle lane and the right-turn reference trajectory. This represents the shortest Euclidean distance between the right boundary of the non-motorized vehicle lane and the right-turn reference trajectory. This is a preset distance threshold, and its value is positive.

[0114] It is understandable that by ensuring the fourth distance meets the aforementioned target range, the target non-motorized vehicle will not collide with surrounding static objects when turning right, such as the curb or the green belt.

[0115] In some implementations, the third distance is matched with each fourth distance separately to obtain multiple second control point information.

[0116] Step S214: If the lane type is a motor vehicle lane, determine the fifth distance based on the first distance, the current speed, and the first duration; determine the second control point information based on the third distance and the fifth distance.

[0117] In some implementations, the fifth distance is determined as shown in the following formula (6): (6); It should be noted that when the lane type is a motor vehicle lane, since the behavior of the target non-motorized vehicle entering the motor vehicle lane does not conform to normal driving habits, only a fifth distance is calculated.

[0118] In some implementations, the third distance is matched with the fifth distance to obtain information on a single second control point.

[0119] In this embodiment, by distinguishing the lane type of the target non-motorized vehicle after turning right and using different processing logics for the second control point information, a single second control point is determined for the motorized vehicle lane scenario, while multiple second control point information is determined for the non-motorized vehicle lane scenario. This takes into account driving rules, generates a second trajectory that conforms to traffic rules more accurately, reduces the generation of invalid second trajectories, and improves the calculation efficiency and accuracy of the second trajectory.

[0120] In some embodiments, step S240 may include steps S241 and S242: Step S241: Determine the first lateral distance between the target object and the right-turn reference trajectory; the target object is located between the motor vehicle lane and the non-motor vehicle lane after the right turn; Here, the target object refers to the object located between the motor vehicle lane and the non-motor vehicle lane, such as a green belt.

[0121] In some implementations, a first lateral distance is determined based on object information of the target object on the map and a right-turn reference trajectory. This first lateral distance characterizes the shortest Euclidean distance between the boundary of the target object on the non-motorized vehicle lane side and the right-turn reference trajectory. Figure 2 As shown .

[0122] Step S242: Determine the lane type based on the first lateral distance and the first distance.

[0123] In some implementations, if the sum of the first lateral distance and the preset target distance is greater than the first distance, it indicates that the target non-motorized vehicle has missed the exit for the non-motorized vehicle lane after turning right. In this scenario, the lane type is determined to be a motorized vehicle lane. Figure 2 As shown in vru3; if the sum of the first lateral distance and the preset target distance is less than or equal to the first distance, it indicates that the target non-motorized vehicle has missed the exit of the non-motorized vehicle lane after turning right. In this scenario, the lane type is determined to be a non-motorized vehicle lane. Figure 2 As shown in vru1 and vru2.

[0124] In this embodiment of the application, based on the first lateral distance and the first distance, the type of lane the target non-motorized vehicle enters after turning right can be identified, thereby providing reasonable prior knowledge for the generation of the second trajectory and reducing the probability of generating a second trajectory that does not conform to traffic rules.

[0125] In some embodiments, step S100 above includes steps S101 and S102: Step S101: If the target intersection between the first direct trajectory and the right-turn reference trajectory is located in the first area, determine the target's traffic intention, including the right-turn intention; the first area refers to the area on the side away from the first stop line from the motor vehicle; the first stop line is the stop line of the motor vehicle lane that the motor vehicle enters after turning right; Here, the target intersection point refers to the intersection between the first direct trajectory and the right-turn reference trajectory. The position of the target intersection point determines the non-motorized vehicle's travel intention. For example... Figure 2 As shown, the intersection of vru2 and the right-turn reference trajectory is located in the first region, which means that the target traffic intention of vru2 is a right-turn intention.

[0126] It should be noted that for the target non-motorized vehicle, its initial intention to travel can be determined as a right turn intention or a cross-traffic intention. Then, based on the relationship between the position of the target intersection point between the first direct trajectory and the right turn reference trajectory and the first stop line, it is determined whether to exclude the cross-traffic intention.

[0127] Step S102: If the target intersection is located in the second area, determine the target's traffic intention, including the intention to turn right and the intention to cross; the second area is located on the side of the first stop line closer to the motor vehicle.

[0128] like Figure 2 As shown, the intersection of vru3 and the right-turn reference trajectory is located in the second region, meaning that the target traffic intention of vru3 is both a right-turn intention and a crossing intention. The intersection of vru1 and the right-turn reference trajectory is also located in the second region, meaning that the target traffic intention of vru1 is both a right-turn intention and a crossing intention.

[0129] In this embodiment, the target travel intention of the non-motorized vehicle is determined by the geometric relationship between the first direct trajectory and the right-turn reference trajectory of the target non-motorized vehicle. This allows for accurate identification of whether the target non-motorized vehicle is only turning right or also intends to cross the road, providing a basis for subsequent trajectory generation and thus improving the rationality and practicality of the prediction.

[0130] In some embodiments, step S100 above includes steps S103 and S104: Step S103: Determine the current position of at least one first non-motorized vehicle; Here, the first non-motorized vehicle refers to any non-motorized vehicle in the current environment of the motorized vehicle. This first non-motorized vehicle can be traveling in the same direction as the motorized vehicle or traveling in the opposite direction. The first non-motorized vehicle can be a bicycle, electric bicycle, or other traffic participant.

[0131] In some implementations, the current position of the first non-motorized vehicle is determined by at least one sensor installed on the motor vehicle, such as lidar, millimeter-wave radar, etc.

[0132] In this application, by identifying the current position of each first non-motorized vehicle, a spatial relationship model between the motorized vehicle and the first non-motorized vehicle is established, and the target non-motorized vehicle is determined through the spatial relationship model.

[0133] Step S104: For each first non-motorized vehicle, determine whether the first non-motorized vehicle is the target non-motorized vehicle based on its current position, target coordinate system, and right-turn reference trajectory; the target coordinate system takes the first intersection point as the origin and the straight direction of the motor vehicle as the x-axis; the first intersection point is the intersection of the right-turn reference trajectory and the stop line of the lane where the motor vehicle is currently located.

[0134] Here, the target coordinate system is a local coordinate system constructed with the first intersection point that the vehicle passes when it is about to turn right as the origin, the straight-ahead direction of the vehicle at its current position defined as the x-axis, and the direction to the left perpendicular to the x-axis as the y-axis. In this application, by introducing the target coordinate system, the complex road environment can be simplified into a unified mathematical model, which facilitates geometric relationship calculations and trajectory fitting.

[0135] In this application, by mapping the current position of the first non-motorized vehicle to the target coordinate system and comparing it with the right-turn reference trajectory, the target non-motorized vehicle can be determined from at least one first non-motorized vehicle. The current lane of the target non-motorized vehicle is parallel to the current lane of the motorized vehicle, and the target non-motorized vehicle and the motorized vehicle are traveling in the same direction in their respective current lanes.

[0136] In this embodiment of the application, by establishing a unified target coordinate system and combining it with a right-turn reference trajectory, the target non-motorized vehicles within the right-turn influence range of motor vehicles can be efficiently identified, providing an input basis for subsequent trajectory prediction.

[0137] In some embodiments, step S104 may include steps S1041 to S1044: Step S1041: Perform vector transformation on the first coordinates of the target point in the target coordinate system to obtain the first vector; In some implementations, the first coordinates of the target point on the right-turn reference trajectory in the target coordinate system are determined; this first coordinate is then converted into a vector representation to obtain a first vector, which can be represented as: .

[0138] Step S1042: Based on the current position of the first non-motorized vehicle, transform it to the second coordinate in the target coordinate system, and determine the second vector; In some implementations, the current position of the first non-motorized vehicle is transformed to the target coordinate system to determine the second coordinates in the target coordinate system; the second coordinates are then converted into a vector representation to obtain the second vector, which can be represented as... .

[0139] Understandably, since the location information of the first non-motorized vehicle may come from different coordinate systems (such as the global coordinate system), it is necessary to map the location information of the first non-motorized vehicle to the target coordinate system in order to maintain consistency with the reference coordinate system of the target point. This process usually involves coordinate transformation matrix operations to ensure that the position of the target point and the current position of the first non-motorized vehicle are comparable in the same coordinate system.

[0140] Step S1043: Based on the first vector and the second vector, determine whether the first non-motorized vehicle is located to the right of the right turn reference trajectory; In some implementations, the relative positional relationship between two vectors is used to determine whether the first non-motorized vehicle is on the right side of the right-turn reference trajectory.

[0141] In practice, the direction of the angle between two vectors is calculated using the cross product of vectors. If the calculation result is negative, the first non-motorized vehicle is determined to be on the right side of the right turn reference trajectory, as shown in the following formula (5).

[0142] (5); The direction of the angle between two vectors is calculated using the cross product of vectors. If the result is positive, the first non-motorized vehicle is determined to be to the left of the right-turn reference trajectory.

[0143] This application determines whether the first non-motorized vehicle is located to the right of the right-turn reference trajectory by using the vector cross product, thereby achieving the first screening of the target non-motorized vehicle among at least one first non-motorized vehicle, effectively reducing the computational amount of trajectory prediction for the target non-motorized vehicle corresponding to the motorized vehicle, and thus improving the computational efficiency of the motorized vehicle.

[0144] In some implementations, if the first non-motorized vehicle is located to the left of the right-turn reference trajectory, it is determined that the first non-motorized vehicle is not the target non-motorized vehicle; if the first non-motorized vehicle is located to the right of the right-turn reference trajectory, step S1044 is executed.

[0145] Step S1044: When the first non-motorized vehicle is located to the right of the right turn reference trajectory, the current heading angle of the first non-motorized vehicle is converted to the target coordinate system to obtain the target heading angle; based on whether the target heading angle meets the preset angle threshold range, it is determined whether the first non-motorized vehicle is the target non-motorized vehicle.

[0146] In some implementations, the angle threshold range may refer to Thus, when the target heading angle is within the angle threshold range, the first non-motorized vehicle is identified as the target non-motorized vehicle. For example, the target heading angle could be 45°. Figure 2 As shown, vru1, vru2, and vru3 are identified as target non-motorized vehicles, while vru4 is not.

[0147] In some implementations, if the target heading angle does not meet the preset angle threshold range, the first non-motorized vehicle is determined to be a non-motorized vehicle.

[0148] In this embodiment of the application, by determining whether the target heading angle is within a reasonable range, it is possible to further verify whether the current driving lane of the first non-motorized vehicle is parallel to the current driving lane of the motorized vehicle, and whether the driving direction of the first non-motorized vehicle is the same as the driving direction of the motorized vehicle, thereby improving the accuracy of the determination of the target non-motorized vehicle.

[0149] In this embodiment, firstly, by performing vector transformation on the target point, it is easier to compare it with the position vector of the first non-motorized vehicle, thereby determining whether the first non-motorized vehicle is located on the right side of the right-turn reference trajectory. Then, if the first non-motorized vehicle is located on the right side of the right-turn reference trajectory, it is further determined whether the first non-motorized vehicle is the target non-motorized vehicle based on its current heading angle. This dual verification method improves the accuracy of determining the target non-motorized vehicle.

[0150] The above-described trajectory prediction method for non-motorized vehicles will be described below with reference to a specific embodiment. For ease of understanding, a possible process applicable to this embodiment will be introduced. However, it is worth noting that this specific embodiment is only for better illustration of this application and does not constitute an improper limitation of this application.

[0151] The navigation-assisted driving system has entered the mass production stage. Compared to main road scenarios, intersection scenarios, due to the lack of lane line constraints, make vehicle interactions more complex and difficult to handle. Specifically, when the vehicle turns right, it faces potential interactions with non-motorized vehicles traveling in the same direction on the right. For example, when a non-motorized vehicle is going straight, a conflict zone will be created, leading to interaction. The throughput and comfort of intersection scenarios have become important indicators for evaluating the capabilities of the navigation-assisted driving system. The predicted trajectory of obstacles, as input to the decision-making and planning module, is a crucial prerequisite for achieving safe interaction and comfortable passage within intersections, significantly impacting the overall functionality.

[0152] However, predicting obstacle trajectories has always been a major challenge in driver assistance systems, especially for non-motorized vehicles, which are more unpredictable and difficult to predict compared to motorized vehicles. Current commonly used trajectory prediction methods are data-driven model-based methods, but these are limited by data quality and distribution, as well as the black-box nature of the data, making long-term performance difficult to guarantee, and kinematic and safety constraints may not be met. Rule-based prediction methods, on the other hand, cannot process a large number of features simultaneously and are easily affected by input from upstream perception modules, resulting in inter-frame jumps in intent. Therefore, a rule-based model-based approach has become an important direction for exploring non-motorized vehicle trajectory prediction, leveraging the feature extraction capabilities of models and the strong constraints of rules to achieve trajectory prediction for non-motorized vehicles.

[0153] Based on this, embodiments of this application provide a trajectory prediction method for a target non-motorized vehicle, including steps S300 to S350: Step S300: Obtain target information about the vehicle and its surrounding environment; Here, the target information of the vehicle and its environment can be determined based on the vehicle's upstream positioning system and perception module. The upstream positioning system can refer to a global navigation satellite system or a real-time dynamic carrier phase differential system, and the perception module can include at least one sensor installed on the vehicle.

[0154] In some implementations, target information may include position, heading angle, rate of change of heading angle, speed, acceleration, and type. Position may refer to the vehicle's position and the position of the target object in its environment; heading angle may refer to the vehicle's heading angle and the heading angle of the target object in its environment; rate of change of heading angle may refer to the rate of change of the vehicle's heading angle and the rate of change of the target object's heading angle in its environment; speed may refer to the vehicle's speed and the speed of the target object in its environment; acceleration may refer to the vehicle's acceleration and the acceleration of the target object in its environment; and type may refer to the type of the vehicle and the type of the target object in its environment. The type of target object may include motor vehicle type, non-motor vehicle type, pedestrian type, curb type, etc.

[0155] In some implementations, the target information may include a reference line for the navigation direction and intersection distance information, wherein the reference line is a reference line (composed of geometric points) provided by an upstream positioning system that conforms to the vehicle's navigation direction; and the intersection distance information may be provided by a light map.

[0156] In some implementations, the starting and ending points of each intersection can be obtained based on reference lines and intersection distance information.

[0157] In some implementations, the road information of the road corresponding to the vehicle's right turn can be determined based on the target information, such as the starting position of the road and the static boundary information of the road (e.g., the boundary information of the green belt and the boundary information of the curb).

[0158] Step S310: Establish a rectangular coordinate system and filter the non-motorized vehicles to be predicted; In some implementations, the direct coordinate system is established as follows: Figure 2 As shown, a rectangular coordinate system is established with the intersection starting point on the reference line as the origin, the straight direction as the x-axis, and the left side perpendicular to the x-axis as the y-axis. The intersection starting point on the reference line can refer to the intersection starting point of the lane where the vehicle is currently located.

[0159] In some implementations, the positions of all first non-motorized vehicles corresponding to the non-motorized type are determined from the target information; based on the positions of all first non-motorized vehicles and a reference line, multiple second non-motorized vehicles are determined from all first non-motorized vehicles; wherein each second non-motorized vehicle is located to the right of the reference line; based on the heading angles of all second non-motorized vehicles, multiple target non-motorized vehicles are determined from all second non-motorized vehicles. The method for determining the second non-motorized vehicles is described below (1), and the method for determining the target non-motorized vehicles is described below (2). (1) Method for determining the second non-motorized vehicle.

[0160] For each non-motorized vehicle, a first reference point is determined from the reference line based on its position, where the Euclidean distance between the first reference point and the first non-motorized vehicle is minimized. The coordinates corresponding to the first reference point are transformed to a Cartesian coordinate system, and a position vector is constructed based on the transformed coordinates. Transform the coordinates corresponding to the position of the first non-motorized vehicle to a rectangular coordinate system, and construct a position vector based on the transformed coordinates. ;exist If the first non-motorized vehicle is located to the right of the reference line, then the first non-motorized vehicle is located to the left of the reference line; otherwise, the first non-motorized vehicle is located to the left of the reference line.

[0161] (2) Method for determining the target non-motorized vehicle.

[0162] For each second non-motorized vehicle, its heading angle is transformed into this rectangular coordinate system to obtain the second heading angle. );like If so, then the second non-motorized vehicle is determined to be the target non-motorized vehicle. Specifically, the target non-motorized vehicle's driving lane is the same as the vehicle's driving lane, and both the target non-motorized vehicle and the vehicle are traveling in the same direction. The second heading angle is the heading angle of the target non-motorized vehicle and... Figure 2 The angle between the x-axis and the x-axis in a Cartesian coordinate system. The angle thresholds (referring to -95° and 30°) can be adjusted according to actual needs.

[0163] It should be noted that the above (2) can be used to filter out the target non-motorized vehicles for trajectory prediction, such as Figure 2 As shown, Figure 2 The target non-motorized vehicles in the text can refer to vru1, vru2, and vru3, while vru4 is not considered a target non-motorized vehicle.

[0164] Step S320: Determine the potential travel intention of the target non-motorized vehicle; In some implementations, the potential travel intentions of multiple target non-motorized vehicles obtained in step S310 are determined separately, wherein the determination of the potential travel intention of a single target non-motorized vehicle is explained.

[0165] During implementation, the initial travel intention of the target non-motorized vehicle is determined, including its intention to cross and its intention to turn right. A direct trajectory curve is derived based on the heading angle of the target non-motorized vehicle. If the intersection of this direct trajectory curve and the reference line is after the intersection of the lane corresponding to the vehicle's right turn, the potential travel intention of the target non-motorized vehicle is determined as a right turn intention. The initial travel intention is determined based on driving experience, categorizing straight and left turns among the possible travel directions of the target non-motorized vehicle as crossing intentions, and right turns among the possible travel directions as right turn intentions.

[0166] Step S330: Determine the set of screening curves for the target non-motorized vehicles; In some implementations, for each target non-motorized vehicle, a prediction curve is determined for each sub-intention among the potential travel intentions of that target non-motorized vehicle; based on the prediction curves of all sub-intentions, a set of curves is selected. The determination of the prediction curve corresponding to each sub-intention is explained below: (1) Prediction curve corresponding to the intention to cross.

[0167] For the intention to cross, a direct trajectory curve is generated based on the heading angle of the target non-motorized vehicle, such as... Figure 2 As shown, vru1 and vru3 each correspond to a direct curve.

[0168] (2) Prediction curve corresponding to the intention to turn right.

[0169] ① Determine the lane type that the target non-motorized vehicle will enter after turning right; During implementation, calculate the lateral distance between the static physical boundary (e.g., green belt) between the motor vehicle lane and the non-motor vehicle lane entered after a vehicle turns right and the reference line. ,like Figure 2 As shown in the figure; then, determine the lateral distance between the position of the target non-motorized vehicle and the reference line. ;exist In this situation, it indicates that the target non-motorized vehicle has missed the exit for the non-motorized vehicle lane. Based on driving experience, if it still turns right, it will most likely enter the motorized vehicle lane. Figure 2 As shown in vru3, otherwise it means that the driver did not miss the non-motorized vehicle exit. Based on driving experience, the driver will most likely enter the non-motorized vehicle lane when turning right. Figure 2 As shown in vru1 and vru2.

[0170] It should be noted that, The preset threshold can be adjusted in practical applications.

[0171] ② Determine the information of the first control point; During implementation, the current location of the target electric vehicle is determined as the information of the first control point.

[0172] ③ Determine the second control point information corresponding to different lane types.

[0173] First, determine the cumulative distance of the second control point on the reference line; Determine the heading difference between the heading angle of the target non-motorized vehicle and the heading angle of the lane after the right turn. (heading); Based on the heading difference and the heading angle of the target non-motorized vehicle, the heading correction time is determined by the above formula (2); the speed of the target non-motorized vehicle is projected onto the reference line to obtain speed conversion information. The position of the target non-motorized vehicle is projected onto the reference line to obtain position conversion information. Based on the above heading correction time, speed conversion information, and position conversion information, the cumulative distance of the second control point on the reference line can be obtained according to the above formula (4). .

[0174] Then, determine the lateral distance between the second control point and the reference line corresponding to different lane types; When turning right into the non-motorized vehicle lane, Using the baseline and the half-width of the non-motorized vehicle as the resolution, multiple different lateral distances are sampled by generalizing left and right using the above formula (4), for example, hour, ; hour, .

[0175] It should be noted that in practical applications, resolution can be balanced between processing time and result completeness. Attention should also be paid to the sampled lateral distance. Need to meet ,in, This is a positive safety threshold, which can be adjusted in practical applications.

[0176] When turning right into the motor vehicle lane, since it does not conform to conventional driving habits, it is sampled only once (meaning only one line is generated). The result can be obtained from the above formula (4). .

[0177] Therefore, the coordinates of the second control point (the position of the end of the turn) on the reference line are: Then, convert them into the corresponding xy coordinates to obtain the coordinate information of the second control point.

[0178] ④ Determine the corresponding third control point information for different second control point information.

[0179] For the third control point (the point after the transition and entering straight-ahead), this embodiment selects it after driving for 1 second based on the second control point. The cumulative distance of the third control point on the reference line can be obtained through the above formula (1). Then the coordinates of the third control point sl are Then, the coordinates are converted into the corresponding xy coordinates to obtain the coordinate information of the third control point.

[0180] ⑤ Based on the coordinate information of the first control point, multiple second control points, and their corresponding third control points, a set of predicted curves corresponding to the right turn intention is generated by Bézier curve fitting.

[0181] (3) Determine the set of screening curves.

[0182] In some implementations, when the potential traffic intentions include both crossing intentions and right-turn intentions, a set of filtering curves is determined based on the prediction curves corresponding to the crossing intentions and the prediction curves corresponding to the right-turn intentions.

[0183] In some implementations, when the potential traffic intention includes a right-turn intention, a set of filter curves is determined based on the prediction curve corresponding to the right-turn intention.

[0184] Step S340: Filter the set of curves to determine the set of candidate trajectories.

[0185] During implementation, the Gaussian decay method is used to extrapolate the current acceleration of the target non-motorized vehicle with a resolution of 0.1s, and the target non-motorized vehicle is assumed to be in uniform acceleration motion within each sampling period.

[0186] First, iterate through each screening curve in the aforementioned set of screening curves. Based on the acceleration and time of each sampling period, determine the distance the target non-motorized vehicle moves within each sampling period. This allows you to find information for each sampling point on the screening curve, thus obtaining a complete 8-second trajectory. Then, determine static obstacle information and boundary information from the target information. Perform static collision detection on the first 4 seconds (an empirical value, adjustable) of each screening trajectory. If a trajectory has a collision, it is discarded. However, it should be noted that at least one solution should be retained for each traffic intention.

[0187] Step S350: Determine the optimal predicted trajectory from the candidate trajectory set.

[0188] During implementation, the target information is input into the prediction model, and the prediction model runs the target information to obtain a set of multimodal predicted trajectories with probabilities. Then, considering that the long-term performance of the multimodal predicted trajectory set with probabilities is not good, the running trajectory of each candidate trajectory in the candidate trajectory set for the first 4 seconds is evaluated with the multimodal predicted trajectory set. The evaluation process can use the average distance error as the trajectory similarity evaluation standard.

[0189] When evaluating each selected curve, the average distance error between the candidate trajectory and each predicted trajectory in the multimodal prediction trajectory set is obtained. Based on all the average distance errors and the corresponding probability information, the weighted average distance error (ADE) of the candidate trajectory is determined. This method can obtain the weighted ADE of each selected trajectory, and thus the candidate trajectory with the smallest weighted ADE is determined as the optimal predicted trajectory for the target non-motorized vehicle.

[0190] This application provides a trajectory prediction method for non-motorized vehicles traveling in the same direction on the right when a vehicle turns right, by combining the advantages of both rule-based and model-based methods. The system or algorithm, considering road environment characteristics, performs targeted sampling and generates candidate trajectories with different travel intentions, thereby avoiding unnecessary sampling results and ensuring that the generated candidate trajectories conform to kinematic laws and meet safety requirements. When evaluating candidate trajectories, the evaluation method uses the similarity between the candidate trajectory and the model output as the evaluation criterion, leveraging the model's feature extraction capabilities to judge the travel intention and the rationality of the candidate trajectory. This avoids the tedious manual parameter adjustment in different scenarios, making the trajectory prediction method proposed in this application highly versatile.

[0191] Based on the foregoing embodiments, this application provides a trajectory prediction device for non-motorized vehicles, such as... Figure 4As shown, the trajectory prediction device 400 for non-motorized vehicles includes: a first determining module 401, used to determine a target non-motorized vehicle with a right-turn intention of a motor vehicle; acquiring the current driving information, current position, and target traffic intention of the target non-motorized vehicle; the current driving information includes the current speed and current heading; a second determining module 402, used to determine a predicted trajectory set of the target non-motorized vehicle based on the current driving information and current position using a target prediction model; a third determining module 403, used to determine a candidate trajectory set based on the right-turn reference trajectory of the motor vehicle, the current driving information, the target traffic intention, and the current position; each candidate trajectory in the candidate trajectory set has no collision risk with a first object in the environment where the motor vehicle is located, excluding the target non-motorized vehicle; and a fourth determining module 404, used to determine the target trajectory from the candidate trajectory set based on the predicted trajectory set and the candidate trajectory set.

[0192] In some embodiments, the third determining module includes: a first determining unit, configured to determine a set of selected trajectories for the target non-motorized vehicle based on a right-turn reference trajectory, current driving information, target traffic intention, and current location; a second determining unit, configured to determine, for each selected trajectory in the set of selected trajectories, the expected acceleration of the target non-motorized vehicle in each sampling period within the sampling duration using a Gaussian decay method based on the current speed; a third determining unit, configured to determine, for each selected trajectory in the set of selected trajectories, the sampling points in each sampling period on the selected trajectory based on the expected acceleration in each sampling period and the current speed; and a fourth determining unit, configured to determine, for each selected trajectory in the set of selected trajectories, a candidate trajectory in the candidate trajectory set based on multiple sampling points within the sampling duration and the location information of the first object.

[0193] In some embodiments, the current heading includes the current heading angle; the target's passage intention includes a crossing intention and a right turn intention; the first determining unit includes: a first determining subunit, configured to determine a first direct trajectory corresponding to the current heading angle for the crossing intention; a second determining subunit, configured to determine at least one second trajectory of the target non-motorized vehicle based on the right turn reference trajectory, the current heading, the current speed, and the current position for the right turn intention; and a third determining subunit, configured to determine a set of filtered trajectories based on at least one second trajectory and the first direct trajectory. In some embodiments, the second determining subunit includes: determining first control point information based on the current position and a right-turn reference trajectory; the first control point information includes a first distance and a second distance; the first distance represents the shortest Euclidean distance between the current position and the right-turn reference trajectory; the second distance represents the distance between the target point on the right-turn reference trajectory corresponding to the shortest Euclidean distance and the starting point of the right-turn reference trajectory; determining at least one second control point based on the first control point information, the current heading, and the current speed; determining third control point information corresponding to the at least one second control point based on the at least one second control point information and the current speed; and determining at least one second trajectory based on the at least one second control point information, the third control point information corresponding to the at least one second control point, the first control point information, and the current position.

[0194] In some embodiments, the apparatus further includes: a fifth determining module, configured to determine the lane type of the target non-motorized vehicle entering the lane after turning right; and a second determining subunit, comprising: determining a first duration based on the current heading and the target heading angle of the target non-motorized vehicle after turning right; determining a third distance based on the first duration, a second distance, and the current speed; determining multiple fourth distances based on the first distance and the width of the target non-motorized vehicle when the lane type is a non-motorized vehicle lane; determining multiple second control point information based on the third distance and the multiple fourth distances; determining a fifth distance based on the first distance, the current speed, and the first duration when the lane type is a motorized vehicle lane; and determining a single second control point information based on the third distance and the fifth distance.

[0195] In some embodiments, the fifth determining module includes: a fifth determining unit, configured to determine a first lateral distance between the target object and the right-turn reference trajectory; the target object is located between the motor vehicle lane and the non-motor vehicle lane after the right turn; and a sixth determining unit, configured to determine the lane type based on the first lateral distance and the first distance.

[0196] In some embodiments, the first determining module further includes: a seventh determining unit, configured to determine that the target's travel intention includes a right-turn intention when the target intersection between the first direct trajectory and the right-turn reference trajectory is located in a first area; the first area refers to the area located on the side away from the vehicle from the first stop line; the first stop line is the stop line of the motor vehicle lane that the vehicle enters after turning right; and an eighth determining unit, configured to determine that the target's travel intention includes a right-turn intention and a crossing intention when the target intersection is located in a second area; the second area is located on the side of the first stop line closer to the vehicle.

[0197] In some embodiments, the first determining module includes: a ninth determining unit, configured to determine the current position corresponding to at least one first non-motorized vehicle; and a tenth determining unit, configured to determine, for each first non-motorized vehicle, whether the first non-motorized vehicle is a target non-motorized vehicle based on the current position of the first non-motorized vehicle, the target coordinate system, and the right-turn reference trajectory; the target coordinate system has the first intersection point as the origin and the straight-going direction of the motor vehicle as the x-axis; the first intersection point is the intersection of the right-turn reference trajectory and the stop line of the lane where the motor vehicle is currently located.

[0198] In some embodiments, the tenth determining unit includes: a first obtaining subunit, configured to perform vector transformation on the first coordinates of the target point in the target coordinate system to obtain a first vector; a fourth determining subunit, configured to transform the current position of the first non-motorized vehicle to the second coordinates in the target coordinate system based on the first and second vectors to determine a second vector; a judging subunit, configured to judge whether the first non-motorized vehicle is located on the right side of the right-turn reference trajectory based on the first and second vectors; and a fifth determining subunit, configured to, if the first non-motorized vehicle is located on the right side of the right-turn reference trajectory, transform the current heading angle of the first non-motorized vehicle to the target coordinate system to obtain a target heading angle; and determine whether the first non-motorized vehicle is a target non-motorized vehicle based on whether the target heading angle meets a preset angle threshold range.

[0199] It should be noted that, in the embodiments of this application, if the above methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of software products. These software products are stored in a storage medium and include several instructions to cause a motor vehicle to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0200] This application also provides a motor vehicle, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the methods described above.

[0201] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. The computer-readable storage medium can be transient or non-transient.

[0202] This application also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, implement some or all of the steps in any of the above-described methods. The computer program product can be implemented specifically through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0203] It should be noted that the descriptions of the storage medium and device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0204] The above embodiments are merely preferred embodiments provided to fully illustrate this application, and the scope of protection of this application is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on this application are all within the scope of protection of this application.

[0205] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for predicting the trajectory of non-motorized vehicles, characterized in that, The method includes: Identify a target non-motorized vehicle that is attempting to turn right; obtain the current driving information, current location, and target traffic intention of the target non-motorized vehicle; the current driving information includes the current speed and current heading; Based on the current driving information and the current location, the target prediction model determines the predicted trajectory set of the target non-motorized vehicle. Based on the right-turn reference trajectory of the motor vehicle, the current driving information, the target traffic intention, and the current location, a candidate trajectory set is determined; each candidate trajectory in the candidate trajectory set has no collision risk with a first object in the environment where the motor vehicle is located, excluding the target non-motorized vehicle; The target trajectory is determined from the candidate trajectory set based on the matching degree between the predicted trajectory set and the candidate trajectory set.

2. The method according to claim 1, characterized in that, The process of determining a candidate trajectory set based on the vehicle's right-turn reference trajectory, the current driving information, the target traffic intention, and the current location includes: Based on the right-turn reference trajectory, the current driving information, the target traffic intention, and the current location, a set of filtered trajectories for the target non-motorized vehicle is determined; For each selected trajectory in the selected trajectory set, based on the current speed, the expected acceleration of the target non-motorized vehicle in each sampling cycle within the sampling time is determined using the Gaussian decay method. For each selected trajectory in the selected trajectory set, based on the expected acceleration and the current velocity in each sampling period, the sampling points in each sampling period on the selected trajectory are determined; For each filtered trajectory in the filtered trajectory set, a candidate trajectory in the candidate trajectory set is determined based on multiple sampling points within the sampling duration and the location information of the first object.

3. The method according to claim 2, characterized in that, The current heading includes the current heading angle; the target passage intention includes a crossing intention and a right turn intention; The process of determining the filter trajectory set for the target non-motorized vehicle based on the right-turn reference trajectory, the current driving information, the target traffic intention, and the current location includes: Based on the stated intention to traverse, determine the first direct trajectory corresponding to the current heading angle; In response to the stated right turn intention, at least one second trajectory for the target non-motorized vehicle is determined based on the right turn reference trajectory, the current heading, the current speed, and the current position. The set of filtered trajectories is determined based on the at least one second trajectory and the first direct trajectory.

4. The method according to claim 2, characterized in that, Determining at least one second trajectory for the target non-motorized vehicle based on the right-turn reference trajectory, the current heading, the current speed, and the current position includes: Based on the current position and the right-turn reference trajectory, first control point information is determined; the first control point information includes a first distance and a second distance; the first distance is used to represent the shortest Euclidean distance between the current position and the right-turn reference trajectory; the second distance is used to represent the distance between the target point corresponding to the shortest Euclidean distance on the right-turn reference trajectory and the starting point of the right-turn reference trajectory. Based on the first control point information, the current heading, and the current speed, at least one second control point information is determined; Based on the at least one second control point information and the current speed, determine the third control point information corresponding to the at least one second control point information respectively; Based on the at least one second control point information, the third control point information corresponding to the at least one second control point information, the first control point information, and the current position, the at least one second trajectory is determined.

5. The method according to claim 4, characterized in that, The method further includes: Determine the lane type that the target non-motorized vehicle will enter after turning right; The step of determining at least one second control point based on the first control point information, the current heading, and the current speed includes: Based on the current heading and the target heading angle after the target non-motorized vehicle turns right, the first duration is determined; Based on the first duration, the second distance, and the current speed, determine the third distance; When the lane type is a non-motorized vehicle lane, multiple fourth distances are determined based on the first distance and the width of the target non-motorized vehicle; multiple second control point information is determined based on the third distance and the multiple fourth distances. If the lane type is a motor vehicle lane, a fifth distance is determined based on the first distance, the current speed, and the first duration; and a single second control point is determined based on the third distance and the fifth distance.

6. The method according to claim 5, characterized in that, Determining the lane type into which the target non-motorized vehicle enters after turning right includes: Determine a first lateral distance between the target object and the right-turn reference trajectory; the target object is located between the motor vehicle lane and the non-motor vehicle lane after the right turn. The lane type is determined based on the first lateral distance and the first distance.

7. The method according to any one of claims 1 to 6, characterized in that, Obtain the target non-motorized vehicle's intended travel, including: If the target intersection point between the first direct trajectory and the right-turn reference trajectory is located in the first area, the target's travel intention is determined to include a right-turn intention; the first area refers to the area located on the side away from the vehicle from the first stop line; the first stop line is the stop line of the vehicle lane that the vehicle enters after making a right turn; If the target intersection is located in the second area, the target traffic intention is determined to include a right turn intention and a crossing intention; the second area is located on the side of the first stop line closer to the vehicle.

8. The method according to any one of claims 1 to 6, characterized in that, The determination of a target non-motorized vehicle when a motor vehicle intends to turn right includes: Determine the current position of at least one first non-motorized vehicle; For each first non-motorized vehicle, based on the current position of the first non-motorized vehicle, the target coordinate system, and the right-turn reference trajectory, it is determined whether the first non-motorized vehicle is the target non-motorized vehicle; the target coordinate system takes the first intersection point as the origin and the straight direction of the motor vehicle as the x-axis; the first intersection point is the intersection of the right-turn reference trajectory and the stop line of the lane where the motor vehicle is currently located.

9. The method according to claim 8, characterized in that, The step of determining whether the first non-motorized vehicle is the target non-motorized vehicle based on the current position of the first non-motorized vehicle, the target coordinate system, and the right-turn reference trajectory includes: Perform a vector transformation on the first coordinates of the target point in the target coordinate system to obtain the first vector; Based on the current position of the first non-motorized vehicle, the second coordinate is determined by transforming it to the second coordinate in the target coordinate system; Based on the first vector and the second vector, determine whether the first non-motorized vehicle is located to the right of the right turn reference trajectory; When the first non-motorized vehicle is located to the right of the right-turn reference trajectory, the current heading angle of the first non-motorized vehicle is converted to the target coordinate system to obtain the target heading angle; based on whether the target heading angle meets the preset angle threshold range, it is determined whether the first non-motorized vehicle is the target non-motorized vehicle.

10. A trajectory prediction device for non-motorized vehicles, characterized in that, The device includes: The first determining module is used to determine a target non-motorized vehicle when the motorized vehicle intends to turn right; and to obtain the current driving information, current position, and target traffic intention of the target non-motorized vehicle; the current driving information includes the current speed and current heading; The second determining module is used to determine the predicted trajectory set of the target non-motorized vehicle based on the current driving information and the current position using a target prediction model; The third determining module is used to determine a set of candidate trajectories based on the right-turn reference trajectory of the motor vehicle, the current driving information, the target traffic intention, and the current position; each candidate trajectory in the set of candidate trajectories has no risk of collision with a first object in the environment where the motor vehicle is located, excluding the target non-motorized vehicle; The fourth determining module is used to determine the target trajectory from the candidate trajectory set based on the matching degree between the predicted trajectory set and the candidate trajectory set.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 9.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

Images