Yaw guiding method, apparatus, electronic device, and computer program product

Abstract

The embodiments of the present disclosure disclose a yaw guiding method and device, electronic equipment and computer program product, the method comprises: obtaining driving data of a navigated object before a target intersection; extracting driving characteristics and a navigation direction of the navigated object before the target intersection based on the driving data; predicting a yaw behavior of the navigated object based on the navigation direction, the driving characteristics and a yaw prediction model corresponding to the target intersection; guiding the navigated object to drive in a correct direction when the yaw behavior of the navigated object is predicted. The technical solution can guide the navigated object to drive into the correct driving direction in time, reduce the yaw probability of the navigated object, and avoid the situation that the navigated object takes a detour.

Description

Technical Field

[0001] This disclosure relates to the field of navigation technology, specifically to a yaw guidance method, device, electronic device, and computer program product. Background Technology

[0002] In travel scenarios, map-based navigation services play a crucial role in navigation applications. Users pay close attention to how to navigate the road ahead, such as how many meters to turn or change lanes. However, while driving, users are prone to distractions, conversations, or other interruptions, which can lead to missing exits and resulting in detours.

[0003] Therefore, a solution is needed to promptly remind users when they show a tendency to deviate from the navigation path and guide them to drive the vehicle in the correct navigation direction, thereby reducing the probability of users deviating from the navigation path. Summary of the Invention

[0004] This disclosure provides a yaw guidance method, apparatus, electronic device, and computer program product.

[0005] In a first aspect, this disclosure provides a yaw guidance method, which includes:

[0006] Obtain the driving data of the navigated object before the target intersection;

[0007] Based on the driving data, the driving characteristics and navigation direction of the navigated object before the target intersection are extracted;

[0008] Based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection, the yaw behavior of the navigated object is predicted;

[0009] When a deviation is predicted in the navigated object, the navigated object is guided to travel in the correct direction.

[0010] Further, the driving data includes the current navigation data of the navigated object before the target intersection and the trajectory data of the navigated object under the navigation of the current navigation data; based on the driving data, the driving characteristics and navigation direction of the navigated object before the target intersection are extracted, including:

[0011] Based on the current navigation data, determine the navigation direction of the sample navigation object before the target intersection;

[0012] Based on the generated trajectory data, determine the correspondence between the speed of the sample navigation object and the distance of the sample navigation object relative to the target intersection.

[0013] Further, the yaw prediction model includes a pre-fitted and trained speed-distance curve model; predicting the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection includes:

[0014] The yaw behavior of the navigated object is determined based on the correspondence between speed and distance and the degree of matching between the speed and distance speed curve models.

[0015] Furthermore, when a veergence is predicted in the navigated object, guiding the navigated object to travel in the correct direction includes:

[0016] Based on the probability of the navigated object deviating from its course, different guidance methods are used to guide the navigated object to travel in the correct navigation direction; the different guidance methods include one or more combinations of different sound effects, different animations, and different voice playback content.

[0017] Furthermore, obtain the driving data of the navigated object before the target intersection, including:

[0018] When the navigated object meets the yaw prediction conditions, the predicted relevant positions before the target intersection included in the yaw prediction model are obtained;

[0019] Based on the predicted relevant location, the driving data of the navigated object is collected at preset time intervals.

[0020] Secondly, this invention provides a yaw prediction model training method, comprising:

[0021] Obtain the driving data of the sample navigation object in front of the target intersection;

[0022] Based on the driving data, driving characteristics of sample navigation objects traveling in different navigation directions before the target intersection are extracted;

[0023] Based on the driving characteristics in different navigation directions, yaw prediction models are trained for different navigation directions so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection.

[0024] Furthermore, based on the driving data, driving characteristics of sample navigation objects traveling in different navigation directions before the target intersection are extracted, including:

[0025] Based on the driving data, determine the navigation direction of the sample navigation object before the target intersection;

[0026] Determine the distribution information of the driving characteristics of the sample navigation object;

[0027] Based on the distribution information, the driving characteristics of the sample navigation objects in the same navigation direction are determined.

[0028] Further, the driving data includes the actual trajectory data generated by the sample navigation object before the target intersection under the navigation of historical navigation data; determining the distribution information of the driving characteristics of the sample navigation object includes:

[0029] Based on the actual trajectory data, determine the speed and lateral offset of the sample navigation object when it travels at different preset distances in front of the target sample intersection;

[0030] Based on the speed and lateral offset of multiple sample navigation objects in the same navigation direction, the distribution information of the speed and lateral offset at different preset distances in the same navigation direction is determined.

[0031] Further, the driving characteristics include the speed and lateral deviation of the sample navigation object at different preset distances during its journey before the target intersection; based on the driving characteristics in different navigation directions, yaw prediction models are trained in different navigation directions so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection, including:

[0032] Based on the speed fitting of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a speed curve model of speed versus distance in different navigation directions is obtained;

[0033] Based on the lateral offset of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a displacement curve model of lateral offset versus distance in different navigation directions is obtained;

[0034] Based on the velocity curve model and displacement curve model, the predicted relevant positions before the target intersection in different navigation directions are determined; wherein, when predicting the yaw behavior of the navigated object, the driving data of the navigated object is collected based on the predicted relevant positions, and the yaw behavior is predicted.

[0035] Furthermore, based on the velocity curve model and displacement curve model, the predicted relevant positions before the target intersection in different navigation directions are determined, including:

[0036] Determine the velocity similarity curves between the velocity curve models in different navigation directions, and determine the displacement similarity curves between the displacement curve models in different navigation directions;

[0037] The first predicted position is determined based on the inflection point on the velocity similarity curve, and the second predicted position is determined based on the inflection point on the displacement similarity curve;

[0038] The predicted relevant location is determined based on the first predicted location and the second predicted location.

[0039] Furthermore, the driving characteristics also include historical yaw information of the sample navigation object; the method further includes:

[0040] Determine the velocity similarity curves between the velocity curve models described in different navigation directions;

[0041] Based on the speed similarity curve, the displacement curve model, and the historical yaw information, it is determined whether the target intersection is suitable for predicting yaw behavior.

[0042] Thirdly, embodiments of the present invention provide a yaw guidance device, comprising:

[0043] The first acquisition module is configured to acquire the driving data of the navigated object before the target intersection;

[0044] The first extraction module is configured to extract the driving characteristics and navigation direction of the navigated object before the target intersection based on the driving data.

[0045] The prediction module is configured to predict the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection;

[0046] The guidance module is configured to guide the navigated object in the correct direction when it is predicted that the navigated object will veer off course.

[0047] Fourthly, this invention provides a yaw prediction model training device, comprising:

[0048] The second acquisition module is configured to acquire driving data of the sample navigation object before the target intersection.

[0049] The second extraction module is configured to extract driving features of sample navigation objects traveling in different navigation directions before the target intersection based on the driving data.

[0050] The training module is configured to train yaw prediction models in different navigation directions based on the driving characteristics in different navigation directions, so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection.

[0051] The function can be implemented in hardware or in hardware with corresponding software. The hardware or software includes one or more modules corresponding to the above function.

[0052] In one possible design, the above-described device includes a memory and a processor. The memory stores one or more computer instructions that support the device in executing the corresponding method described above, and the processor is configured to execute the computer instructions stored in the memory. The device may also include a communication interface for communicating with other devices or communication networks.

[0053] Fifthly, embodiments of this disclosure provide an electronic device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in any of the preceding aspects.

[0054] In a sixth aspect, embodiments of this disclosure provide a computer-readable storage medium for storing computer instructions used by any of the above-described devices, which, when executed by a processor, are used to implement the methods described in any of the above aspects.

[0055] In a seventh aspect, embodiments of this disclosure provide a computer program product comprising computer instructions which, when executed by a processor, are used to implement the methods described in any of the preceding aspects.

[0056] The technical solutions provided in this disclosure may have the following beneficial effects:

[0057] This embodiment of the disclosure targets a navigated object traveling at a target intersection. It acquires the navigated object's driving data before the intersection, extracts its driving characteristics and navigation direction from the data, and then predicts the navigated object's deviation behavior based on the navigation direction, driving characteristics, and a pre-trained yaw prediction model corresponding to the target intersection. When yaw behavior is predicted, the navigated object is guided back to the correct driving direction. Through this method, when the navigated object deviates from its course due to its own reasons or external influences during navigation, it can be promptly guided back to the correct driving direction, reducing the probability of deviation and preventing the navigated object from taking detours.

[0058] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0059] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments, taken in conjunction with the accompanying drawings. In the drawings:

[0060] Figure 1 A flowchart illustrating a yaw guidance method according to an embodiment of the present disclosure is shown;

[0061] Figure 2A flowchart is shown for a yaw prediction model training method according to an embodiment of the present disclosure;

[0062] Figure 3 A schematic diagram showing the selection of inflection points on a transverse curve according to an embodiment of the present disclosure is provided.

[0063] Figure 4 This diagram illustrates the lateral offset range in different navigation directions according to an embodiment of the present disclosure.

[0064] Figure 5A A schematic diagram showing the median curve of the lateral offset interval in different navigation directions according to an embodiment of the present disclosure;

[0065] Figure 5B A schematic diagram of displacement similarity curves according to an embodiment of the present disclosure is shown;

[0066] Figure 6 A structural block diagram of a yaw guidance device according to an embodiment of the present disclosure is shown;

[0067] Figure 7 A structural block diagram of a yaw prediction model training apparatus according to an embodiment of the present disclosure is shown.

[0068] Figure 8 This is a schematic diagram of the structure of an electronic device suitable for implementing the yaw guidance method and / or yaw prediction model training method according to an embodiment of the present disclosure. Detailed Implementation

[0069] In the following, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings to enable those skilled in the art to readily implement them. Furthermore, for clarity, portions unrelated to the description of the exemplary embodiments have been omitted from the drawings.

[0070] In this disclosure, it should be understood that terms such as “comprising” or “having” are intended to indicate the presence of features, figures, steps, behaviors, components, parts or combinations thereof disclosed in this specification, and do not preclude the possibility of the presence or addition of one or more other features, figures, steps, behaviors, components, parts or combinations thereof.

[0071] It should also be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0072] The details of the embodiments of this disclosure are described in detail below through specific examples.

[0073] Figure 1 A flowchart illustrating a yaw guidance method according to an embodiment of the present disclosure is shown. Figure 1As shown, the yaw guidance method includes the following steps:

[0074] In step S101, the driving data of the navigated object before the target intersection is obtained;

[0075] In step S102, the driving characteristics and navigation direction of the navigated object before the target intersection are extracted based on the driving data;

[0076] In step S103, the yaw behavior of the navigated object is predicted based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection;

[0077] In step S104, when it is predicted that the navigable object will veer off course, the navigable object is guided to travel in the correct direction.

[0078] In this embodiment, the yaw guidance method can be executed on a navigation terminal or a navigation server. The navigated object can be any object currently traveling before the target intersection, such as a user, an autonomous vehicle, or an intelligent robot. The target intersection can be any intersection for which a yaw prediction model has been pre-trained. The intersection can be predefined in the road network data, typically referring to an intersection with branching roads.

[0079] This embodiment performs yaw prediction for a navigated object traveling before a target intersection. During yaw prediction, the navigated object's driving data before the target intersection can be acquired. This driving data may include, but is not limited to, the navigated object's actual driving trajectory (acquired with the navigated object's authorization) and navigation data output from the navigation service to the navigated object.

[0080] After obtaining the driving data of the navigated object, its driving characteristics and navigation direction can be extracted based on the data. Driving characteristics may include, but are not limited to, the speed of the navigated object as it travels towards the target intersection, its distance from the target intersection, and its lateral deviation from the road before the target intersection. If the road driving direction is understood as vertical, then the lateral deviation can be understood as the amount of deviation of the navigated object in the lateral direction of the road, such as the amount of deviation of the navigated object relative to a certain edge line or center line of the road. The navigation direction can be the direction the navigated object should travel before the target intersection, that is, the correct driving direction of the navigated object before the target intersection. The navigation direction may include, but is not limited to, exiting to the right, exiting to the left, or going straight.

[0081] In some embodiments, a corresponding yaw prediction model is pre-trained for the target intersection. The yaw prediction model can predict the yaw behavior of objects traveling in different navigation directions before the target intersection. In some embodiments, different yaw prediction models can be trained for different navigation directions. The corresponding yaw prediction model can be selected based on the navigation direction of the navigated object. Then, the yaw prediction model can be used to predict whether the navigated object has yaw behavior based on the driving characteristics of the navigated object. For example, if the navigation direction of the navigated object should be to exit to the right, but the yaw prediction model corresponding to exiting to the right determines based on the driving characteristics of the navigated object that the navigated object has not actually exited to the right, or the possibility of exiting to the right is small, then the yaw behavior of the navigated object can be predicted.

[0082] In some embodiments, the presence of yaw behavior in the navigated object can be determined by matching the degree of matching between the driving characteristics of the navigated object and the yaw prediction model.

[0083] In some embodiments, driving features may also include, but are not limited to, features such as the speed and / or lateral deviation of the navigated object under different road conditions before the target intersection (e.g., road grade, road structure, number of lanes, navigation segment length, mixed roads, etc.) and different environments (e.g., weather conditions, time of day, traffic congestion, etc.). It should be noted that the speed and / or lateral deviation features of the navigated object in the driving features correspond to the relative distance between the navigated object and the target intersection.

[0084] In some embodiments, the lateral offset of the navigated object can be determined based on the perpendicular distance between the navigated object at a preset distance before the target intersection and the centerline of the road before the target intersection. The edge line of the road before the target intersection is known, and the centerline can be determined based on the coordinates of various points on the edge line. The position of the navigated object at the preset distance can also be determined based on trajectory data. Therefore, the lateral offset of the navigated object can be determined based on the position of the navigated object at the preset distance and the position coordinates of the centerline of the road before the target intersection.

[0085] In some embodiments, the yaw prediction model can be a fitted curve model between speed and distance, that is, the yaw prediction model can be a curve model based on the speed of the sample navigation object before the target intersection and the distance between the sample navigation object and the target intersection. Of course, the yaw prediction model can also be a more complex model, such as a model trained by learning multiple features of the sample navigation object at different distances before the target intersection, such as speed, heading angle, and lateral offset, like a neural network model. The yaw prediction model can also be trained separately based on different road conditions and / or different road environments, that is, different road conditions and / or different environments can correspond to different yaw prediction models. In the actual prediction process, the appropriate yaw prediction model can be selected for prediction based on the current road conditions and road environment. For example, different yaw prediction models can be trained for the same target intersection in different navigation directions, and different target intersections can also correspond to different yaw prediction models; of course, multiple target intersections with similar or identical road conditions can correspond to the same yaw prediction model in the same navigation direction.

[0086] This embodiment of the disclosure targets a navigated object traveling at a target intersection. It acquires the navigated object's driving data before the intersection, extracts its driving characteristics and navigation direction from the data, and then predicts the navigated object's deviation behavior based on the navigation direction, driving characteristics, and a pre-trained yaw prediction model corresponding to the target intersection. When yaw behavior is predicted, the navigated object is guided back to the correct driving direction. Through this method, when the navigated object deviates from its course due to its own reasons or external influences during navigation, it can be promptly guided back to the correct driving direction, reducing the probability of deviation and preventing the navigated object from taking detours.

[0087] In an optional implementation of this embodiment, the driving data includes the current navigation data of the navigated object before the target intersection and the trajectory data of the navigated object under the navigation of the current navigation data; step S102, namely the step of extracting the driving characteristics and navigation direction of the navigated object before the target intersection based on the driving data, further includes the following steps:

[0088] Based on the current navigation data, determine the navigation direction of the sample navigation object before the target intersection;

[0089] Based on the generated trajectory data, determine the correspondence between the speed of the sample navigation object and the distance of the sample navigation object relative to the target intersection.

[0090] In this optional implementation, yaw prediction is performed during the navigation process of the navigated object. Therefore, the driving data of the navigated object obtained may include, but is not limited to, the current navigation data output by the navigation server to the navigated object, and may also include trajectory data generated by the navigated object over a period of time. The current navigation data may include, but is not limited to, directions indicating the direction the navigated object should travel, such as exiting on the right, exiting on the left, or going straight.

[0091] As the navigated object travels towards the target intersection, it uses the current navigation data to drive. During this process, trajectory data is generated, which may include, but is not limited to, the speed of the navigated object at each trajectory point. Based on the location information of each trajectory point and the location information of the target intersection, the distance between the navigated object at each trajectory point and the target intersection can be determined. In turn, the correspondence between the speed of the navigated object and the distance between the navigated object and the target intersection can be determined.

[0092] In some embodiments, the velocity at each trajectory point generated by the navigated object can be acquired at preset time intervals, for example, the velocity at one trajectory point per second. In order to accurately predict the yaw behavior of the navigated object, after acquiring a sufficient number of velocities at trajectory points, the yaw behavior of the navigated object can be predicted using a yaw prediction model based on the velocities at these sufficient trajectory points and their corresponding distances.

[0093] In an optional implementation of this embodiment, the yaw prediction model includes a speed curve model of speed versus distance obtained through pre-fitting and training; step S103, which is the step of predicting the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection, further includes the following steps:

[0094] The yaw behavior of the navigated object is determined based on the correspondence between speed and distance and the degree of matching between the speed and distance speed curve models.

[0095] In this optional implementation, a speed curve model of speed versus distance in different navigation directions can be pre-trained based on the driving characteristics of the sample navigation object before the target intersection. This speed curve model represents the speed curve of the sample navigation object at various preset distances before the target intersection. During actual navigation, after the correspondence between the speed and distance of the navigated object before the target intersection is determined, the speed curve model of speed versus distance corresponding to the correct driving direction of the navigated object (i.e., the navigation direction indicated in the current navigation data) can be matched based on this correspondence. If the matching degree is high, the possibility of the navigated object deviating from its course is considered low; if the matching degree is low, the possibility of the navigated object deviating from its course is considered high. In other embodiments, the speed curve model of speed versus distance corresponding to the incorrect driving direction of the navigated object (i.e., the navigation direction not indicated in the current navigation data) can also be matched based on this correspondence. If the matching degree is high, the possibility of the navigated object deviating from its course is considered high; if the matching degree is low, the possibility of the navigated object deviating from its course is considered low. In other embodiments, the first matching degree of the speed curve model corresponding to the correct driving direction and the second matching degree of the speed curve model corresponding to the wrong driving direction can be considered together. For example, the final matching degree can be obtained by weighting the first matching degree and the second matching degree, and the velocity behavior of the navigation behavior can be determined based on the final matching degree.

[0096] In an optional implementation of this embodiment, step S104, which is the step of guiding the navigated object to travel in the correct direction when it is predicted that the navigated object will veer off course, further includes the following steps:

[0097] Based on the probability of the navigated object deviating from its course, different guidance methods are used to guide the navigated object to travel in the correct direction; the different guidance methods include one or more combinations of different sound effects, different animations, and different voice playback content.

[0098] In this optional implementation, the yaw prediction model can predict the probability of the navigated object exhibiting yaw behavior. This embodiment can employ a differentiated guidance method, using different guidance methods for different probability levels. The guidance method can include, but is not limited to, one or more combinations of different sound effects, different animations, and different voice playback content.

[0099] Multiple sound effects, animations, and voice playback content can be preset. Based on the probability predicted by the current yaw prediction model, one or more combinations of sound effects, animations, and voice playback content can be selected to remind the navigated object of its current yaw behavior.

[0100] The following example illustrates a differentiated guidance approach:

[0101]

[0102] In an optional implementation of this embodiment, step S101, namely the step of obtaining the driving data of the navigated object before the target intersection, further includes the following steps:

[0103] When the navigated object meets the yaw prediction conditions, the predicted relevant position before the target intersection determined in the yaw prediction model is obtained;

[0104] Based on the predicted relevant location, the driving data of the navigated object is collected at preset time intervals.

[0105] In this optional implementation, yaw prediction conditions can be preset. For example, it can be set whether the navigation segment before the current target intersection is suitable for yaw prediction (in the prediction stage, this can be determined based on the similarity of the driving characteristics of sample navigation objects in different navigation directions. If the driving characteristics of sample navigation objects before a target intersection in different navigation directions are similar, it is impossible to accurately distinguish whether the driving object before the target intersection is yawed, then the navigation segment before the target intersection is not suitable for yaw prediction; if the driving characteristics of sample navigation objects before the target intersection in different navigation directions are dissimilar in some cases, it is possible to accurately distinguish the yaw behavior of the driving object before the target intersection based on this dissimilarity, then the navigation segment before the target intersection is suitable for yaw prediction), and whether the distance between the navigated object and the navigation segment of the target intersection is greater than a certain distance (if the distance is too short, it is not enough to collect enough driving data for prediction).

[0106] Under the condition of satisfying the yaw prediction, a pre-determined prediction-related position corresponding to the target intersection can be obtained. This prediction-related position is determined when training the yaw prediction model. In some embodiments, the prediction-related position may include, but is not limited to, the start collection position, the start prediction position, and the end prediction position. The start collection position can be a position where driving data collection can begin, the start prediction position can be a position where yaw behavior can begin to be predicted, and the end prediction position can be a position where prediction needs to end. In some embodiments, the prediction-related position can be expressed as the distance of the navigated object relative to the target intersection, such as how many meters away from the target intersection constitutes the start collection position, the start prediction position, or the end prediction position.

[0107] Different yaw prediction models define different prediction-related locations before the target intersection. When it is determined that yaw prediction is required, these prediction-related locations can be obtained first, and then the driving data of the navigated object can be collected at preset time intervals based on these locations. For example, after the navigated object reaches the starting collection position, the collection of its driving data begins. After reaching the starting prediction position, driving features are extracted based on the previously collected driving data, and the yaw behavior of the navigated object is predicted based on the yaw prediction model. After reaching the ending prediction position, the collection of driving data and the prediction of yaw behavior can be stopped.

[0108] Figure 2 A flowchart illustrating a yaw prediction model training method according to an embodiment of this disclosure is shown. Figure 2 As shown, the training method for this yaw prediction model includes the following steps:

[0109] In step S201, the driving data of the sample navigation object before the target intersection is obtained;

[0110] In step S202, driving features of sample navigation objects traveling in different navigation directions before the target intersection are extracted based on the driving data;

[0111] In step S203, a yaw prediction model for different navigation directions is trained based on the driving characteristics in different navigation directions, so that the yaw prediction model can predict the yaw behavior of the navigated object before the target intersection.

[0112] In this embodiment, the yaw prediction model training method can be executed on a navigation server. The sample navigation object can be any object using navigation services to navigate before the target intersection within a preset time range, such as a user, a self-driving vehicle, or a smart robot. The target intersection can be any intersection where the yaw prediction model needs to be trained. In some embodiments, during the yaw prediction model training process, driving data can be acquired for a batch of target intersections, for example, driving data can be acquired for all intersections in a city or a specific area of ​​a city, and then the driving data corresponding to each intersection can be grouped.

[0113] For each target intersection, driving data from multiple sample navigation objects within a preset time range can be obtained. In some embodiments, the driving data may include, but is not limited to, driving-related data generated by the sample navigation objects traveling within a preset distance before the target intersection within the preset time range, such as navigation data, trajectory data, etc. That is to say, the driving data may include, but is not limited to, the actual driving trajectory of the sample navigation objects within a preset distance before the target intersection (obtained with the authorization of the sample navigation objects), navigation data output by the navigation service to the sample navigation objects, etc.

[0114] After obtaining the driving data of the sample navigation objects, their driving features can be extracted based on this data. It's important to note that when extracting these features, the driving features of sample navigation objects traveling in different navigation directions can be distinguished. In other words, sample navigation objects can be grouped according to their navigation direction. Then, sample navigation objects traveling in the same direction will have their driving features extracted from that direction, while sample navigation objects traveling in different directions will have their driving features extracted from different groups. It's understood that the navigation direction mentioned here refers to the navigation direction indicated by the navigation data before the target intersection.

[0115] In some embodiments, sample navigation objects can be classified into two categories—normal driving and eccentric driving—based on whether the actual driving direction and the navigation direction are the same. If the actual driving direction and the navigation direction of a sample navigation object are consistent during navigation before the target intersection, the sample navigation object can be considered a normally driving object. If the actual driving direction and the navigation direction are inconsistent, the sample navigation object can be considered an eccentric driving object. For example, during navigation, sample navigation object A is prompted by navigation data to exit to the right before the target intersection, but sample navigation object A does not actually exit to the right and continues straight, thus missing the target intersection. In this case, sample navigation object A can be classified as an eccentric driving sample navigation object. Similarly, during navigation, sample navigation object B is prompted by navigation data to exit to the right before the target intersection, and sample navigation object B also exits to the right and from the target intersection. In this case, sample navigation object B can be classified as a normally driving sample navigation object. The driving characteristics of normally driving and eccentric driving sample navigation objects can be distinguished for subsequent model training.

[0116] In some embodiments, driving characteristics may include, but are not limited to, the speed of the sample navigation object as it travels before the target intersection, the distance to the target intersection, and the lateral offset on the road before the target intersection. If the road driving direction is understood as vertical, the lateral offset can be understood as the amount of offset of the sample navigation object in the lateral direction of the road, such as the amount of offset of the sample navigation object relative to a certain edge line or center line of the road. The navigation direction can be the direction that the sample navigation object should travel when navigating before the target intersection, that is, the correct driving direction indicated by the navigation data before the target intersection. The navigation direction may include, but is not limited to, exiting to the right, exiting to the left, or going straight.

[0117] In this embodiment, a yaw prediction model is trained using the extracted driving characteristics of sample navigation objects. This model can predict the yaw behavior of the navigated object traveling in different navigation directions before the target intersection. In some embodiments, different yaw prediction models can be trained for different navigation directions. During online prediction, the corresponding yaw prediction model can be selected based on the navigation direction of the sample navigation object. Then, the yaw prediction model can be used to predict whether the sample navigation object has yaw behavior based on the driving characteristics of the navigated object. For example, if the navigation direction of the navigated object should be to exit to the right, but the yaw prediction model for exiting to the right determines based on the driving characteristics of the navigated object that the navigated object has not actually exited to the right, or the probability of exiting to the right is small, then the yaw behavior of the navigated object can be predicted.

[0118] In some embodiments, the yaw prediction model may include, but is not limited to, a curve model between speed and distance, which may be a curve obtained by fitting the speed and distance relative to the target intersection in the driving characteristics of the sample navigation object.

[0119] For example, this curve model can be expressed as a curve function in the following form:

[0120] v = ax 4 +bx 3 +cx 2 +dx+e

[0121] Where v represents velocity, x represents distance, and a, b, c, d, and e represent the model parameters to be trained.

[0122] Of course, it's understandable that yaw prediction models can be more complex. For example, they could be models trained by learning multiple features such as the speed, heading angle, and lateral offset of a sample navigation object at different distances before the target intersection, like neural network models. Yaw prediction models can also be trained separately based on different road conditions and / or road environments. That is, different road conditions and / or different environments can correspond to different yaw prediction models, and in actual prediction, the appropriate yaw prediction model can be selected based on the current road conditions and environment. For example, different yaw prediction models can be trained for the same target intersection on different navigation directions, and different target intersections can also correspond to different yaw prediction models. Of course, multiple target intersections with similar or identical road conditions can correspond to the same yaw prediction model on the same navigation direction. No specific restrictions are imposed here; it can be determined based on actual needs.

[0123] In some embodiments, driving features may also include, but are not limited to, features such as the speed and / or lateral deviation of the navigated object under different road conditions before the target intersection (e.g., road grade, road structure, number of lanes, navigation segment length, mixed roads, etc.) and different environments (e.g., weather conditions, time of day, road congestion, etc.). It should be noted that the speed and / or lateral deviation features of the navigated object in the driving features correspond to the relative distance between the navigated object and the target intersection. The above driving features can be used to train a yaw prediction model so that, during online prediction, the yaw prediction model can also predict whether the navigated object exhibits yaw behavior based on the above driving features. It is understood that driving features are not limited to speed, lateral deviation, etc., mentioned above, and may also include other features, such as heading angle, yaw rate, etc.

[0124] This embodiment of the disclosure, for each target intersection, collects driving data of sample navigation objects navigating before the target intersection within a preset time range. It then extracts driving characteristics of the sample navigation objects driving normally and abnormally in different navigation directions before the target intersection from the driving data. Based on these driving characteristics in different navigation directions, it trains yaw prediction models for different navigation directions, enabling these models to predict the yaw behavior of the navigated object before the target intersection. Through this method, yaw prediction models can be trained based on the driving characteristics of sample navigation objects. Therefore, when the navigated object deviates from its course due to its own reasons or external influences during navigation, it can be promptly guided back to the correct driving direction, reducing the probability of the navigated object's deviation and preventing it from taking detours.

[0125] In an optional implementation of this embodiment, step S202, namely the step of extracting the driving characteristics of sample navigation objects traveling in different navigation directions before the target intersection based on the driving data, further includes the following steps:

[0126] Based on the driving data, determine the navigation direction of the sample navigation object before the target intersection;

[0127] Determine the driving characteristics of the sample navigation objects in the same navigation direction and the distribution information of the driving characteristics.

[0128] In this optional implementation, the navigation direction of the sample navigation object before the target intersection can be determined first using driving data. Based on the navigation direction, the sample navigation objects are then divided into different groups, with the same navigation direction corresponding to the same group of sample navigation objects, and different navigation directions corresponding to different groups of sample navigation objects. As mentioned above, the navigation direction of the sample navigation object is the correct driving direction of the sample navigation object before it passes the target intersection, which can be determined, for example, through navigation data output by the navigation service.

[0129] After grouping the sample navigation objects according to navigation direction, the distribution information of driving characteristics can be extracted for each group of sample navigation objects, and then the driving characteristics of that group of sample navigation objects can be determined based on this distribution information. In other words, the driving characteristics of sample navigation objects in the same navigation direction are a comprehensive expression of the driving characteristics of that group of sample navigation objects.

[0130] In some embodiments, for each sample navigation object in each group, its corresponding driving features can be extracted first. Then, the distribution information of the sample navigation objects in the navigation direction can be determined based on the driving features of the group of sample navigation objects. This distribution information can be, for example, hotspot distribution information. If the driving feature is a speed feature, the hotspot distribution of the speeds of each sample navigation object in the same navigation direction at multiple preset distances relative to the target intersection can be determined using the hotspot distribution information. Based on the hotspot distribution, the speed features of the group of sample navigation objects, that is, the speed features of the sample navigation objects in the navigation direction, can be determined. In this way, the driving features of multiple sample navigation objects in the same navigation direction can be comprehensively considered, thereby finding a more accurate driving feature that can characterize the sample navigation objects in the navigation direction.

[0131] In some embodiments, the driving characteristics may include, but are not limited to, one or more combinations of speed, yaw information, and lateral offset of the sample navigation object under different scenarios. Different scenarios may include, but are not limited to, scenarios formed by one or more combinations of different road conditions and different environments. The speed, lateral offset, and yaw information of the sample navigation object can be obtained through the actual trajectory data generated by the sample navigation object during navigation and the navigation data output by the navigation server. Yaw information may include information on whether the sample navigation object veers off course when navigating before the target intersection.

[0132] In an optional implementation of this embodiment, the driving data includes historical navigation data of the sample navigation object before the target intersection; the step of determining the navigation direction of the sample navigation object before the target intersection based on the driving data further includes the following steps:

[0133] The navigation direction of the sample navigation object before the target intersection is determined based on the historical navigation data.

[0134] In this optional implementation, the sample navigation object is an object that navigates past the target intersection within a preset time range. Therefore, the driving data of the sample navigation object may include, but is not limited to, historical navigation data output by the navigation server to the sample navigation object within the preset time range. Historical navigation data may include, but is not limited to, directions indicating the direction the sample navigation object should travel, such as exiting to the right, exiting to the left, or going straight. Therefore, the navigation direction of the sample navigation object before the target intersection can be obtained from the historical navigation data.

[0135] In an optional implementation of this embodiment, the driving data includes the actual trajectory data generated by the sample navigation object before the target intersection under the navigation of historical navigation data; the step of determining the distribution information of the driving characteristics of the sample navigation object further includes the following steps:

[0136] Based on the actual trajectory data, determine the speed and lateral offset of the sample navigation object when it travels at different preset distances in front of the target sample intersection;

[0137] Based on the speed and lateral offset of multiple sample navigation objects in the same navigation direction, the distribution information of the speed and lateral offset at different preset distances in the same navigation direction is determined.

[0138] In this optional implementation, as described above, the sample navigation objects include multiple objects. The driving characteristics of each sample navigation object before the target intersection can be determined based on its actual trajectory data, such as speed and lateral offset. This actual trajectory data may include, but is not limited to, the position information and speed of the sample navigation object at each trajectory point. Based on the position information of each trajectory point and the position information of the target intersection, the distance between the navigated object at each trajectory point and the target intersection can be determined. Furthermore, the correspondence between the speed of the sample navigation object and its distance relative to the target intersection can be established.

[0139] For multiple sample navigation objects in the same navigation direction, the distribution information of the sample navigation objects in that navigation direction can be determined based on their respective driving characteristics such as speed and lateral offset; such distribution information can be, for example, hotspot distribution information.

[0140] In an optional implementation of this embodiment, the driving characteristics include the speed and lateral deviation of the sample navigation object at different preset distances during its travel before the target intersection; the step of training yaw prediction models in different navigation directions based on the driving characteristics in different navigation directions, so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection, further includes the following steps:

[0141] Based on the speed fitting of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a speed curve model of speed versus distance in different navigation directions is obtained;

[0142] Based on the lateral offset of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a displacement curve model of lateral offset versus distance in different navigation directions is obtained;

[0143] Based on the velocity curve model and displacement curve model, the predicted relevant positions before the target intersection in different navigation directions are determined; wherein, when predicting the yaw behavior of the navigated object, the driving data of the navigated object is collected based on the predicted relevant positions, and the yaw behavior is predicted.

[0144] In this optional implementation, yaw prediction models can be trained separately for different navigation directions before the target intersection. The same navigation direction at the same target intersection can correspond to the same yaw prediction model, while different navigation directions at the same target intersection can correspond to different yaw prediction models.

[0145] In some embodiments, during the training of the yaw prediction model, driving features can be extracted based on the driving data of the sample navigation data. These driving features may include, but are not limited to, the speed and lateral offset of the sample navigation object at different preset distances during its navigation process before the target intersection. Different preset distances refer to the distances of the sample navigation object relative to the target intersection. In some embodiments, driving data within a preset distance range, such as 1 kilometer, relative to the target intersection can be collected throughout the day. The preset distances can be positions at intervals within that preset distance range, for example, setting a preset distance every 20 meters starting from 1 kilometer. In this way, multiple speed and lateral offset information at multiple preset distances can be extracted from the driving data.

[0146] For the velocities of multiple sample navigation objects along the same navigation direction at various preset distances, the final velocity at each preset distance can be determined based on the velocity distribution information. The final velocity at each preset distance can be identified by using the velocity distribution information of the sample navigation objects at that preset distance to best represent the velocity of the majority of sample navigation objects at that preset distance. The velocities and lateral offsets at various preset distances along all navigation directions can be obtained using the same method.

[0147] A velocity curve model in the same navigation direction can be obtained by fitting the velocity at each preset distance in the same navigation direction, and a displacement curve model in the same navigation direction can be obtained by fitting the lateral offset at each preset distance in the same direction.

[0148] Furthermore, when predicting yaw behavior online, driving data can be collected starting from a pre-defined prediction-related position, and yaw behavior prediction can begin. This prediction-related position can be determined during the training of the yaw prediction model based on the velocity curve model and the displacement curve model; different navigation directions can correspond to different prediction-related positions.

[0149] In some embodiments, the prediction-related location may include, but is not limited to, a start collection location, a start prediction location, and an end prediction location. The start collection location may be a location where driving data collection can begin, the start prediction location may be a location where veergence behavior can begin to be predicted, and the end prediction location may be a location where prediction needs to end. In some embodiments, the prediction-related location may be expressed as the distance of the sample navigation object relative to the target intersection, such as how many meters away from the target intersection constitutes the start collection location, the start prediction location, or the end prediction location.

[0150] In some embodiments, the predicted relevant position before the target intersection can be defined in the yaw prediction model for non-navigation directions. When predicting yaw behavior, the driving data of the navigated object can be collected based on this predicted relevant position. For example, after the navigated object reaches the starting collection position, the collection of the navigated object's driving data begins. After reaching the starting prediction position, driving features are extracted based on the previously collected driving data, and the yaw behavior of the navigated object is predicted based on the yaw prediction model. After reaching the ending prediction position, the collection of driving data and the prediction of yaw behavior can be stopped.

[0151] The lateral offset of the sample navigation object can be determined based on the perpendicular distance between the sample navigation object and the centerline of the road in front of the target intersection at a preset distance. The edge line of the road in front of the target intersection is known, and the centerline can be determined based on the coordinates of various points on the edge line. The position of the sample navigation object at the preset distance can also be determined based on trajectory data. Therefore, the lateral offset can be determined based on the position of the sample navigation object at the preset distance and the position coordinates of the centerline of the road in front of the target intersection.

[0152] In an optional implementation of this embodiment, the step of determining the predicted relevant position before the target intersection based on the velocity curve model and the displacement curve model in different navigation directions further includes the following steps:

[0153] Determine the velocity similarity curves between the velocity curve models in different navigation directions, and determine the displacement similarity curves between the displacement curve models in different navigation directions;

[0154] The first predicted position is determined based on the inflection point on the velocity similarity curve, and the second predicted position is determined based on the inflection point on the displacement similarity curve;

[0155] The predicted relevant location is determined based on the first predicted location and the second predicted location.

[0156] In this optional implementation, corresponding velocity curve models and displacement curve models can be trained for the target intersection in different navigation directions. To compare the characteristics of sample navigation objects traveling in different navigation directions in terms of velocity and lateral offset, the similarity between velocity curve models and displacement curve models in different navigation directions can be compared.

[0157] For example, different navigation directions include going straight and exiting to the right. We can determine the speed similarity curve between the speed curve model corresponding to going straight and the speed curve model corresponding to exiting to the right. Then, based on the speed similarity curve, we can determine the similarities and differences in speed between the two types of navigation objects, going straight and exiting to the right. Similarly, we can obtain the displacement similarity curve. Then, based on the displacement similarity curve, we can compare the similarities and differences in lateral offset between the two types of navigation objects, going straight and exiting to the right.

[0158] Inflection points on velocity and displacement similarity curves can be understood as points where navigation objects in different navigation directions change from similar to dissimilar in their driving characteristics. These inflection points are represented as maxima or minima on the curves. Therefore, predicted relevant locations can be determined based on these inflection points.

[0159] The following example illustrates a method for determining and predicting relevant locations based on inflection points. We will use going straight and driving to the right as examples. It is understood that different navigation directions are not limited to going straight and driving to the right.

[0160] The set of inflection points can be found based on the speed similarity curves of going straight and exiting to the right. An inflection point is recorded as the first predicted position when it meets the following conditions:

[0161] 1. There is an inflection point within 200-1000 meters relative to the target intersection, and it is an inflection point (peak) of a downward trend. The slope of the inflection point is greater than the preset slope (e.g., 0.3). The calculation method of 0.3 is to take 100 meters as a unit value on the horizontal axis. It means that within 100 meters, the change in similarity must exceed 0.3 to meet the requirement, that is, (y2-y1) / ((x2-x1) / 100)>0.3, where (x1,y1) and (x2,y2) are two adjacent points on the speed similarity curve.

[0162] 2. The number of inflection points in the inflection point set is greater than 1 and less than 3 (if there are too many inflection points, it indicates that the velocity fluctuation is too large and it is not suitable as a prediction location).

[0163] 3. Select the inflection point with the steepest slope (significantly different similarities) from the set of inflection points as the final inflection point. The x-coordinate of this final inflection point is then used as the first predicted position. Figure 3 The inflection point at 400 meters shown was selected as the final inflection point.

[0164] The selection principle for the second predicted position is similar to that for the first predicted position, except that the second predicted position is based on the displacement similarity curve, which will not be elaborated here.

[0165] The following example illustrates one method for determining the lateral offset and obtaining the inflection point on the displacement curve model:

[0166] In this example, we will still use going straight and exiting on the right as examples to illustrate how lateral offset is determined. Lateral offset for different navigation directions can be set as an interval, such as... Figure 4 As shown, the two directions, namely execution and driving to the right, correspond to two sections, with four curves.

[0167] The lateral offset interval is obtained based on the position coordinate data of the sample navigation object in the same direction as the target intersection. The horizontal axis can be the distance relative to the target intersection, and the vertical axis can be the difference between the position coordinate point of the sample navigation object and the position coordinate point of the center line of the road in front of the target intersection.

[0168] The median curve for each direction is calculated based on the two boundary curves of the interval corresponding to each direction, resulting in the following: Figure 5A The two curves shown, one for going straight and the other for exiting to the right, are displacement curve models in the directions of going straight and exiting to the right.

[0169] Calculate the difference between the two displacement curve models in the straight and right-hand exit directions to obtain the displacement similarity curve, as shown below. Figure 5B As shown.

[0170] for Figure 5B The principle of the inflection point calculation and filtering method shown above is the same as the description of the velocity similarity curve. The difference is that the inflection point of the upward trend (i.e., the bottom point) is taken for the displacement curve model, rather than the inflection point of the downward trend.

[0171] Based on the first and second predicted positions, a prediction reference position can be calculated. For example, the prediction reference position can be the one that is farther away from the target intersection than the first word position or the second predicted position.

[0172] In some embodiments, predicting relevant locations may include, but is not limited to, starting to predict locations, starting to collect locations, and / or stopping to predict locations.

[0173] After determining the prediction baseline location, the prediction related locations can be determined based on the prediction baseline location.

[0174] For example, the starting prediction position is a location a certain distance above the prediction reference position, meaning the starting prediction position is farther from the target intersection than the prediction reference position. This allows prediction to begin a few seconds earlier. The starting data collection position can be a location a certain distance above the starting prediction position, again to collect driving data a few seconds before the starting prediction position. This ensures sufficient driving data has been collected at the starting prediction position to extract driving features for use in the yaw prediction model. The ending prediction position can be set at a preset distance from the target intersection, such as 200 meters away.

[0175] In an optional implementation of this embodiment, the driving characteristics further include the historical yaw information of the sample navigation object; the method further includes the following steps:

[0176] Determine the velocity similarity curves between the velocity curve models described in different navigation directions;

[0177] Based on the speed similarity curve, the displacement curve model, and the historical yaw information, it is determined whether the target intersection is suitable for predicting yaw behavior.

[0178] In this optional implementation, after obtaining the velocity curve model during model training, the similarity between velocity curve models in different navigation directions can be calculated. If the different velocity curve models are relatively similar, the navigated object before the target intersection may be difficult to distinguish its actual driving direction before the target intersection using the velocity curve model. In this case, the displacement curve model and historical yaw information can be used for further confirmation. If the displacement curve models of the navigated objects in different navigation directions are also relatively similar, and / or the historical yaw rate of the navigated object before the target intersection is low, then the target intersection can be considered unsuitable for yaw behavior prediction, and the target intersection can be marked as an intersection where yaw behavior prediction is not performed. When the navigated object actually uses the navigation service, it does not need to perform yaw behavior prediction when passing through the target intersection. For target intersections that do not meet the above conditions, the target intersection can be marked as suitable for yaw behavior prediction, and the navigated object can perform yaw behavior prediction when passing through the target intersection when actually using the navigation service. It should be noted that the historical deviation information of a sample navigation object can be a comprehensive expression of the historical deviation information of multiple sample navigation objects before the target intersection, such as the average deviation rate of multiple sample navigation objects.

[0179] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein.

[0180] Figure 6 A structural block diagram of a yaw guidance device according to an embodiment of the present disclosure is shown. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 6 As shown, the yaw guidance device includes:

[0181] The first acquisition module 601 is configured to acquire the driving data of the navigated object before the target intersection;

[0182] The first extraction module 602 is configured to extract the driving characteristics and navigation direction of the navigated object before the target intersection based on the driving data.

[0183] Prediction module 603 is configured to predict the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection;

[0184] The guidance module 604 is configured to guide the navigated object to travel in the correct direction when it is predicted that the navigated object will veer off course.

[0185] In this embodiment, the yaw guidance device can be executed on a navigation terminal or a navigation server. The object being navigated can be any object currently traveling before the target intersection, such as a user, an autonomous vehicle, or an intelligent robot. The target intersection can be any intersection for which a yaw prediction model has been pre-trained. The intersection can be predefined in the road network data, typically referring to an intersection with branching roads.

[0186] This embodiment performs yaw prediction for a navigated object traveling before a target intersection. During yaw prediction, the navigated object's driving data before the target intersection can be acquired. This driving data may include, but is not limited to, the navigated object's actual driving trajectory (acquired with the navigated object's authorization) and navigation data output from the navigation service to the navigated object.

[0187] After obtaining the driving data of the navigated object, its driving characteristics and navigation direction can be extracted based on the data. Driving characteristics may include, but are not limited to, the speed of the navigated object as it travels towards the target intersection, its distance from the target intersection, and its lateral deviation from the road before the target intersection. If the road driving direction is understood as vertical, then the lateral deviation can be understood as the amount of deviation of the navigated object in the lateral direction of the road, such as the amount of deviation of the navigated object relative to a certain edge line or center line of the road. The navigation direction can be the direction the navigated object should travel before the target intersection, that is, the correct driving direction of the navigated object before the target intersection. The navigation direction may include, but is not limited to, exiting to the right, exiting to the left, or going straight.

[0188] In some embodiments, a corresponding yaw prediction model is pre-trained for the target intersection. The yaw prediction model can predict the yaw behavior of objects traveling in different navigation directions before the target intersection. In some embodiments, different yaw prediction models can be trained for different navigation directions. The corresponding yaw prediction model can be selected based on the navigation direction of the navigated object. Then, the yaw prediction model can be used to predict whether the navigated object has yaw behavior based on the driving characteristics of the navigated object. For example, if the navigation direction of the navigated object should be to exit to the right, but the yaw prediction model corresponding to exiting to the right determines based on the driving characteristics of the navigated object that the navigated object has not actually exited to the right, or the possibility of exiting to the right is small, then the yaw behavior of the navigated object can be predicted.

[0189] In some embodiments, the presence of yaw behavior in the navigated object can be determined by matching the degree of matching between the driving characteristics of the navigated object and the yaw prediction model.

[0190] In some embodiments, driving features may also include, but are not limited to, features such as the speed and / or lateral deviation of the navigated object under different road conditions before the target intersection (e.g., road grade, road structure, number of lanes, navigation segment length, mixed roads, etc.) and different environments (e.g., weather conditions, time of day, traffic congestion, etc.). It should be noted that the speed and / or lateral deviation features of the navigated object in the driving features correspond to the relative distance between the navigated object and the target intersection.

[0191] In some embodiments, the lateral offset of the navigated object can be determined based on the perpendicular distance between the navigated object at a preset distance before the target intersection and the centerline of the road before the target intersection. The edge line of the road before the target intersection is known, and the centerline can be determined based on the coordinates of various points on the edge line. The position of the navigated object at the preset distance can also be determined based on trajectory data. Therefore, the lateral offset of the navigated object can be determined based on the position of the navigated object at the preset distance and the position coordinates of the centerline of the road before the target intersection.

[0192] In some embodiments, the yaw prediction model can be a fitted curve model between speed and distance, that is, the yaw prediction model can be a curve model based on the speed of the sample navigation object before the target intersection and the distance between the sample navigation object and the target intersection. Of course, the yaw prediction model can also be a more complex model, such as a model trained by learning multiple features of the sample navigation object at different distances before the target intersection, such as speed, heading angle, and lateral offset, like a neural network model. The yaw prediction model can also be trained separately based on different road conditions and / or different road environments, that is, different road conditions and / or different environments can correspond to different yaw prediction models. In the actual prediction process, the appropriate yaw prediction model can be selected for prediction based on the current road conditions and road environment. For example, different yaw prediction models can be trained for the same target intersection in different navigation directions, and different target intersections can also correspond to different yaw prediction models; of course, multiple target intersections with similar or identical road conditions can correspond to the same yaw prediction model in the same navigation direction.

[0193] This embodiment of the disclosure targets a navigated object traveling at a target intersection. It acquires the navigated object's driving data before the intersection, extracts its driving characteristics and navigation direction from the data, and then predicts the navigated object's deviation behavior based on the navigation direction, driving characteristics, and a pre-trained yaw prediction model corresponding to the target intersection. When yaw behavior is predicted, the navigated object is guided back to the correct driving direction. Through this method, when the navigated object deviates from its course due to its own reasons or external influences during navigation, it can be promptly guided back to the correct driving direction, reducing the probability of deviation and preventing the navigated object from taking detours.

[0194] In an optional implementation of this embodiment, the driving data includes the current navigation data of the navigated object before the target intersection and the trajectory data of the navigated object generated under the navigation of the current navigation data; the first extraction module includes:

[0195] The first determining submodule is configured to determine the navigation direction of the sample navigation object before the target intersection based on the current navigation data;

[0196] The second determining submodule is configured to determine the correspondence between the speed of the sample navigation object and the distance of the sample navigation object relative to the target intersection based on the generated trajectory data.

[0197] In this optional implementation, yaw prediction is performed during the navigation process of the navigated object. Therefore, the driving data of the navigated object obtained may include, but is not limited to, the current navigation data output by the navigation server to the navigated object, and may also include trajectory data generated by the navigated object over a period of time. The current navigation data may include, but is not limited to, directions indicating the direction the navigated object should travel, such as exiting on the right, exiting on the left, or going straight.

[0198] As the navigated object travels towards the target intersection, it uses the current navigation data to drive. During this process, trajectory data is generated, which may include, but is not limited to, the speed of the navigated object at each trajectory point. Based on the location information of each trajectory point and the location information of the target intersection, the distance between the navigated object at each trajectory point and the target intersection can be determined. In turn, the correspondence between the speed of the navigated object and the distance between the navigated object and the target intersection can be determined.

[0199] In some embodiments, the velocity at each trajectory point generated by the navigated object can be acquired at preset time intervals, for example, the velocity at one trajectory point per second. In order to accurately predict the yaw behavior of the navigated object, after acquiring a sufficient number of velocities at trajectory points, the yaw behavior of the navigated object can be predicted using a yaw prediction model based on the velocities at these sufficient trajectory points and their corresponding distances.

[0200] In an optional implementation of this embodiment, the yaw prediction model includes a pre-fitted and trained velocity-distance curve model; the prediction module includes:

[0201] The third determining submodule is configured to determine the yaw behavior of the navigated object based on the correspondence between speed and distance and the degree of matching between the speed and distance speed curve models.

[0202] In this optional implementation, a speed curve model of speed versus distance in different navigation directions can be pre-trained based on the driving characteristics of the sample navigation object before the target intersection. This speed curve model represents the speed curve of the sample navigation object at various preset distances before the target intersection. During actual navigation, after the correspondence between the speed and distance of the navigated object before the target intersection is determined, the speed curve model of speed versus distance corresponding to the correct driving direction of the navigated object (i.e., the navigation direction indicated in the current navigation data) can be matched based on this correspondence. If the matching degree is high, the possibility of the navigated object deviating from its course is considered low; if the matching degree is low, the possibility of the navigated object deviating from its course is considered high. In other embodiments, the speed curve model of speed versus distance corresponding to the incorrect driving direction of the navigated object (i.e., the navigation direction not indicated in the current navigation data) can also be matched based on this correspondence. If the matching degree is high, the possibility of the navigated object deviating from its course is considered high; if the matching degree is low, the possibility of the navigated object deviating from its course is considered low. In other embodiments, the first matching degree of the speed curve model corresponding to the correct driving direction and the second matching degree of the speed curve model corresponding to the wrong driving direction can be considered together. For example, the final matching degree can be obtained by weighting the first matching degree and the second matching degree, and the velocity behavior of the navigation behavior can be determined based on the final matching degree.

[0203] In an optional implementation of this embodiment, the guiding module includes:

[0204] The guidance submodule is configured to guide the navigated object to the correct navigation direction using different guidance methods based on the probability of the navigated object deviating from its course; the different guidance methods include one or more combinations of different sound effects, different animations, and different voice playback content.

[0205] In this optional implementation, the yaw prediction model can predict the probability of the navigated object exhibiting yaw behavior. This embodiment can employ a differentiated guidance method, using different guidance methods for different probability levels. The guidance method can include, but is not limited to, one or more combinations of different sound effects, different animations, and different voice playback content.

[0206] Multiple sound effects, animations, and voice playback content can be preset. Based on the probability predicted by the current yaw prediction model, one or more combinations of sound effects, animations, and voice playback content can be selected to remind the navigated object of its current yaw behavior.

[0207] In an optional implementation of this embodiment, the first acquisition module includes:

[0208] The first acquisition submodule is configured to acquire the predicted relevant position before the target intersection included in the yaw prediction model when the navigated object meets the yaw prediction condition.

[0209] The data acquisition submodule is configured to acquire the driving data of the navigated object at preset time intervals based on the predicted relevant location.

[0210] In this optional implementation, yaw prediction conditions can be preset. For example, it can be set whether the navigation segment before the current target intersection is suitable for yaw prediction (in the prediction stage, this can be determined based on the similarity of the driving characteristics of sample navigation objects in different navigation directions. If the driving characteristics of sample navigation objects before a target intersection in different navigation directions are similar, it is impossible to accurately distinguish whether the driving object before the target intersection is yawed, then the navigation segment before the target intersection is not suitable for yaw prediction; if the driving characteristics of sample navigation objects before the target intersection in different navigation directions are dissimilar in some cases, it is possible to accurately distinguish the yaw behavior of the driving object before the target intersection based on this dissimilarity, then the navigation segment before the target intersection is suitable for yaw prediction), and whether the distance between the navigated object and the navigation segment of the target intersection is greater than a certain distance (if the distance is too short, it is not enough to collect enough driving data for prediction).

[0211] Under the condition of satisfying the yaw prediction, a pre-determined prediction-related position corresponding to the target intersection can be obtained. This prediction-related position is determined when training the yaw prediction model. In some embodiments, the prediction-related position may include, but is not limited to, the start collection position, the start prediction position, and the end prediction position. The start collection position can be a position where driving data collection can begin, the start prediction position can be a position where yaw behavior can begin to be predicted, and the end prediction position can be a position where prediction needs to end. In some embodiments, the prediction-related position can be expressed as the distance of the navigated object relative to the target intersection, such as how many meters away from the target intersection constitutes the start collection position, the start prediction position, or the end prediction position.

[0212] Different yaw prediction models define different prediction-related locations before the target intersection. When it is determined that yaw prediction is required, these prediction-related locations can be obtained first, and then the driving data of the navigated object can be collected at preset time intervals based on these locations. For example, after the navigated object reaches the starting collection position, the collection of its driving data begins. After reaching the starting prediction position, driving features are extracted based on the previously collected driving data, and the yaw behavior of the navigated object is predicted based on the yaw prediction model. After reaching the ending prediction position, the collection of driving data and the prediction of yaw behavior can be stopped.

[0213] Figure 7 A structural block diagram of a yaw prediction model training apparatus according to an embodiment of the present disclosure is shown. This apparatus can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 7 As shown, the yaw prediction model training device includes:

[0214] The second acquisition module 701 is configured to acquire driving data of the sample navigation object before the target intersection.

[0215] The second extraction module 702 is configured to extract driving features of sample navigation objects traveling in different navigation directions before the target intersection based on the driving data;

[0216] Training module 703 is configured to train yaw prediction models in different navigation directions based on the driving characteristics in different navigation directions, so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection.

[0217] In this embodiment, the yaw prediction model training device can be executed on a navigation server. The sample navigation object can be any object navigating using navigation services before the target intersection within a preset time range, such as a user, a self-driving vehicle, or a smart robot. The target intersection can be any intersection where the yaw prediction model needs to be trained. In some embodiments, during the yaw prediction model training process, driving data can be acquired for a batch of target intersections, for example, driving data can be acquired for all intersections in a city or a specific area of ​​a city, and then the driving data corresponding to each intersection can be grouped.

[0218] For each target intersection, driving data from multiple sample navigation objects within a preset time range can be obtained. In some embodiments, the driving data may include, but is not limited to, driving-related data generated by the sample navigation objects traveling within a preset distance before the target intersection within the preset time range, such as navigation data, trajectory data, etc. That is to say, the driving data may include, but is not limited to, the actual driving trajectory of the sample navigation objects within a preset distance before the target intersection (obtained with the authorization of the sample navigation objects), navigation data output by the navigation service to the sample navigation objects, etc.

[0219] After obtaining the driving data of the sample navigation objects, their driving features can be extracted based on this data. It's important to note that when extracting these features, the driving features of sample navigation objects traveling in different navigation directions can be distinguished. In other words, sample navigation objects can be grouped according to their navigation direction. Then, sample navigation objects traveling in the same direction will have their driving features extracted from that direction, while sample navigation objects traveling in different directions will have their driving features extracted from different groups. It's understood that the navigation direction mentioned here refers to the navigation direction indicated by the navigation data before the target intersection.

[0220] In some embodiments, sample navigation objects can be classified into two categories—normal driving and eccentric driving—based on whether the actual driving direction and the navigation direction are the same. If the actual driving direction and the navigation direction of a sample navigation object are consistent during navigation before the target intersection, the sample navigation object can be considered a normally driving object. If the actual driving direction and the navigation direction are inconsistent, the sample navigation object can be considered an eccentric driving object. For example, during navigation, sample navigation object A is prompted by navigation data to exit to the right before the target intersection, but sample navigation object A does not actually exit to the right and continues straight, thus missing the target intersection. In this case, sample navigation object A can be classified as an eccentric driving sample navigation object. Similarly, during navigation, sample navigation object B is prompted by navigation data to exit to the right before the target intersection, and sample navigation object B also exits to the right and from the target intersection. In this case, sample navigation object B can be classified as a normally driving sample navigation object. The driving characteristics of normally driving and eccentric driving sample navigation objects can be distinguished for subsequent model training.

[0221] In some embodiments, driving characteristics may include, but are not limited to, the speed of the sample navigation object during its journey before the target intersection, the distance to the target intersection, and the lateral deviation on the road before the target intersection. The navigation direction may be the direction that the sample navigation object should travel when navigating before the target intersection, that is, the correct driving direction indicated by the navigation data of the sample navigation object before the target intersection. The navigation direction may include, but is not limited to, exiting on the right, exiting on the left, or going straight.

[0222] In this embodiment, a yaw prediction model is trained using the extracted driving characteristics of sample navigation objects. This model can predict the yaw behavior of the navigated object traveling in different navigation directions before the target intersection. In some embodiments, different yaw prediction models can be trained for different navigation directions. During online prediction, the corresponding yaw prediction model can be selected based on the navigation direction of the sample navigation object. Then, the yaw prediction model can be used to predict whether the sample navigation object has yaw behavior based on the driving characteristics of the navigated object. For example, if the navigation direction of the navigated object should be to exit to the right, but the yaw prediction model for exiting to the right determines based on the driving characteristics of the navigated object that the navigated object has not actually exited to the right, or the probability of exiting to the right is small, then the yaw behavior of the navigated object can be predicted.

[0223] In some embodiments, the yaw prediction model may include, but is not limited to, a curve model between speed and distance, which may be a curve obtained by fitting the speed and distance relative to the target intersection in the driving characteristics of the sample navigation object.

[0224] Of course, it's understandable that yaw prediction models can be more complex. For example, they could be models trained by learning multiple features such as the speed, heading angle, and lateral offset of a sample navigation object at different distances before the target intersection, like neural network models. Yaw prediction models can also be trained separately based on different road conditions and / or road environments. That is, different road conditions and / or different environments can correspond to different yaw prediction models, and in actual prediction, the appropriate yaw prediction model can be selected based on the current road conditions and environment. For example, different yaw prediction models can be trained for the same target intersection on different navigation directions, and different target intersections can also correspond to different yaw prediction models. Of course, multiple target intersections with similar or identical road conditions can correspond to the same yaw prediction model on the same navigation direction. No specific restrictions are imposed here; it can be determined based on actual needs.

[0225] In some embodiments, driving features may also include, but are not limited to, features such as the speed and / or lateral deviation of the navigated object under different road conditions before the target intersection (e.g., road grade, road structure, number of lanes, navigation segment length, mixed roads, etc.) and different environments (e.g., weather conditions, time of day, road congestion, etc.). It should be noted that the speed and / or lateral deviation features of the navigated object in the driving features correspond to the relative distance between the navigated object and the target intersection. The above driving features can be used to train a yaw prediction model so that, during online prediction, the yaw prediction model can also predict whether the navigated object exhibits yaw behavior based on the above driving features. It is understood that driving features are not limited to speed, lateral deviation, etc., mentioned above, and may also include other features, such as heading angle, yaw rate, etc.

[0226] This embodiment of the disclosure, for each target intersection, collects driving data of sample navigation objects navigating before the target intersection within a preset time range. It then extracts driving characteristics of the sample navigation objects driving normally and abnormally in different navigation directions before the target intersection from the driving data. Based on these driving characteristics in different navigation directions, it trains yaw prediction models for different navigation directions, enabling these models to predict the yaw behavior of the navigated object before the target intersection. Through this method, yaw prediction models can be trained based on the driving characteristics of sample navigation objects. Therefore, when the navigated object deviates from its course due to its own reasons or external influences during navigation, it can be promptly guided back to the correct driving direction, reducing the probability of the navigated object's deviation and preventing it from taking detours.

[0227] In an optional implementation of this embodiment, the second extraction module includes:

[0228] The fourth determining submodule is configured to determine the navigation direction of the sample navigation object before the target intersection based on the driving data;

[0229] The fifth determining submodule is configured to determine the distribution information of the driving characteristics of the sample navigation object;

[0230] The sixth determining submodule is configured to determine the driving characteristics of the sample navigation objects in the same navigation direction based on the distribution information.

[0231] In this optional implementation, the navigation direction of the sample navigation object before the target intersection can be determined first using driving data. Based on the navigation direction, the sample navigation objects are then divided into different groups, with the same navigation direction corresponding to the same group of sample navigation objects, and different navigation directions corresponding to different groups of sample navigation objects. As mentioned above, the navigation direction of the sample navigation object is the correct driving direction of the sample navigation object before it passes the target intersection, which can be determined, for example, through navigation data output by the navigation service.

[0232] After grouping the sample navigation objects according to navigation direction, the distribution information of driving characteristics can be extracted for each group of sample navigation objects, and then the driving characteristics of that group of sample navigation objects can be determined based on this distribution information. In other words, the driving characteristics of sample navigation objects in the same navigation direction are a comprehensive expression of the driving characteristics of that group of sample navigation objects.

[0233] In some embodiments, for each sample navigation object in each group, its corresponding driving features can be extracted first. Then, the distribution information of the sample navigation objects in the navigation direction can be determined based on the driving features of the group of sample navigation objects. This distribution information can be, for example, hotspot distribution information. If the driving feature is a speed feature, the hotspot distribution of the speeds of each sample navigation object in the same navigation direction at multiple preset distances relative to the target intersection can be determined using the hotspot distribution information. Based on the hotspot distribution, the speed features of the group of sample navigation objects, that is, the speed features of the sample navigation objects in the navigation direction, can be determined. In this way, the driving features of multiple sample navigation objects in the same navigation direction can be comprehensively considered, thereby finding a more accurate driving feature that can characterize the sample navigation objects in the navigation direction.

[0234] In some embodiments, the driving characteristics may include, but are not limited to, one or more combinations of speed, yaw information, and lateral offset of the sample navigation object under different scenarios. Different scenarios may include, but are not limited to, scenarios formed by one or more combinations of different road conditions and different environments. The speed, lateral offset, and yaw information of the sample navigation object can be obtained through the actual trajectory data generated by the sample navigation object during navigation and the navigation data output by the navigation server. Yaw information may include information on whether the sample navigation object veers off course when navigating before the target intersection.

[0235] In an optional implementation of this embodiment, the driving data includes the historical navigation data of the sample navigation object before the target intersection; the fourth determining submodule is implemented as follows:

[0236] The navigation direction of the sample navigation object before the target intersection is determined based on the historical navigation data.

[0237] In this optional implementation, the sample navigation object is an object that navigates past the target intersection within a preset time range. Therefore, the driving data of the sample navigation object may include, but is not limited to, historical navigation data output by the navigation server to the sample navigation object within the preset time range. Historical navigation data may include, but is not limited to, directions indicating the direction the sample navigation object should travel, such as exiting to the right, exiting to the left, or going straight. Therefore, the navigation direction of the sample navigation object before the target intersection can be obtained from the historical navigation data.

[0238] In an optional implementation of this embodiment, the driving data includes the actual trajectory data generated by the sample navigation object before the target intersection under the navigation of historical navigation data; the fifth determining submodule includes:

[0239] The seventh determination submodule is configured to determine the speed and lateral offset of the sample navigation object when it travels at different preset distances in front of the target sample intersection based on the actual trajectory data;

[0240] The eighth determining submodule is configured to determine the distribution information of the speed and the lateral offset at different preset distances in the same navigation direction based on the speed and the lateral offset corresponding to multiple sample navigation objects in the same navigation direction.

[0241] In this optional implementation, as described above, the sample navigation objects include multiple objects. The driving characteristics of each sample navigation object before the target intersection can be determined based on its actual trajectory data, such as speed and lateral offset. This actual trajectory data may include, but is not limited to, the position information and speed of the sample navigation object at each trajectory point. Based on the position information of each trajectory point and the position information of the target intersection, the distance between the navigated object at each trajectory point and the target intersection can be determined. Furthermore, the correspondence between the speed of the sample navigation object and its distance relative to the target intersection can be established.

[0242] For multiple sample navigation objects in the same navigation direction, the distribution information of the sample navigation objects in that navigation direction can be determined based on their respective driving characteristics such as speed and lateral offset; such distribution information can be, for example, hotspot distribution information.

[0243] In an optional implementation of this embodiment, the driving features include the speed and lateral offset of the sample navigation object at different preset distances during its journey before the target intersection; the training module includes:

[0244] The first fitting submodule is configured to fit the velocity curve model of the sample navigation object in different navigation directions at different preset distances in front of the target intersection to obtain the velocity curve model of the velocity versus distance in different navigation directions.

[0245] The second fitting submodule is configured to fit the displacement curve model of lateral offset versus distance in different navigation directions based on the lateral offset of the sample navigation object at different preset distances in front of the target intersection in different navigation directions;

[0246] The ninth determining submodule is configured to determine the predicted relevant position before the target intersection in different navigation directions based on the velocity curve model and the displacement curve model; wherein, when predicting the yaw behavior of the navigated object, the driving data of the navigated object is collected based on the predicted relevant position, and the yaw behavior is predicted.

[0247] In this optional implementation, yaw prediction models can be trained separately for different navigation directions before the target intersection. The same navigation direction at the same target intersection can correspond to the same yaw prediction model, while different navigation directions at the same target intersection can correspond to different yaw prediction models.

[0248] In some embodiments, during the training of the yaw prediction model, driving features can be extracted based on the driving data of the sample navigation data. These driving features may include, but are not limited to, the speed and lateral offset of the sample navigation object at different preset distances during its navigation process before the target intersection. Different preset distances refer to the distances of the sample navigation object relative to the target intersection. In some embodiments, driving data within a preset distance range, such as 1 kilometer, relative to the target intersection can be collected throughout the day. The preset distances can be positions at intervals within that preset distance range, for example, setting a preset distance every 20 meters starting from 1 kilometer. In this way, multiple speed and lateral offset information at multiple preset distances can be extracted from the driving data.

[0249] For the velocities of multiple sample navigation objects along the same navigation direction at various preset distances, the final velocity at each preset distance can be determined based on the velocity distribution information. The final velocity at each preset distance can be identified by using the velocity distribution information of the sample navigation objects at that preset distance to best represent the velocity of the majority of sample navigation objects at that preset distance. The velocities and lateral offsets at various preset distances along all navigation directions can be obtained using the same method.

[0250] A velocity curve model in the same navigation direction can be obtained by fitting the velocity at each preset distance in the same navigation direction, and a displacement curve model in the same navigation direction can be obtained by fitting the lateral offset at each preset distance in the same direction.

[0251] Furthermore, when predicting yaw behavior online, driving data can be collected starting from a pre-defined prediction-related position, and yaw behavior prediction can begin. This prediction-related position can be determined during the training of the yaw prediction model based on the velocity curve model and the displacement curve model; different navigation directions can correspond to different prediction-related positions.

[0252] In some embodiments, the prediction-related location may include, but is not limited to, a start collection location, a start prediction location, and an end prediction location. The start collection location may be a location where driving data collection can begin, the start prediction location may be a location where veergence behavior can begin to be predicted, and the end prediction location may be a location where prediction needs to end. In some embodiments, the prediction-related location may be expressed as the distance of the sample navigation object relative to the target intersection, such as how many meters away from the target intersection constitutes the start collection location, the start prediction location, or the end prediction location.

[0253] In some embodiments, the predicted relevant position before the target intersection can be defined in the yaw prediction model for non-navigation directions. When predicting yaw behavior, the driving data of the navigated object can be collected based on this predicted relevant position. For example, after the navigated object reaches the starting collection position, the collection of the navigated object's driving data begins. After reaching the starting prediction position, driving features are extracted based on the previously collected driving data, and the yaw behavior of the navigated object is predicted based on the yaw prediction model. After reaching the ending prediction position, the collection of driving data and the prediction of yaw behavior can be stopped.

[0254] The lateral offset of the sample navigation object can be determined based on the perpendicular distance between the sample navigation object and the centerline of the road in front of the target intersection at a preset distance. The edge line of the road in front of the target intersection is known, and the centerline can be determined based on the coordinates of various points on the edge line. The position of the sample navigation object at the preset distance can also be determined based on trajectory data. Therefore, the lateral offset can be determined based on the position of the sample navigation object at the preset distance and the position coordinates of the centerline of the road in front of the target intersection.

[0255] In an optional implementation of this embodiment, the ninth determining submodule includes:

[0256] The tenth determining submodule is configured to determine the velocity similarity curves between the velocity curve models in different navigation directions, and to determine the displacement similarity curves between the displacement curve models in different navigation directions;

[0257] The eleventh determination submodule is configured to determine a first predicted position based on the inflection point on the velocity similarity curve, and to determine a second predicted position based on the inflection point on the displacement similarity curve.

[0258] The twelfth determination submodule is configured to determine the prediction related position based on the first prediction position and the second prediction position.

[0259] In this optional implementation, corresponding velocity curve models and displacement curve models can be trained for the target intersection in different navigation directions. To compare the characteristics of sample navigation objects traveling in different navigation directions in terms of velocity and lateral offset, the similarity between velocity curve models and displacement curve models in different navigation directions can be compared.

[0260] For example, different navigation directions include going straight and exiting to the right. We can determine the speed similarity curve between the speed curve model corresponding to going straight and the speed curve model corresponding to exiting to the right. Then, based on the speed similarity curve, we can determine the similarities and differences in speed between the two types of navigation objects, going straight and exiting to the right. Similarly, we can obtain the displacement similarity curve. Then, based on the displacement similarity curve, we can compare the similarities and differences in lateral offset between the two types of navigation objects, going straight and exiting to the right.

[0261] Inflection points on velocity and displacement similarity curves can be understood as location nodes where navigation objects in different navigation directions change from similar to dissimilar in their driving characteristics. Therefore, predicted relevant locations can be determined based on these inflection points.

[0262] Based on the first and second predicted positions, a prediction reference position can be calculated. For example, the prediction reference position can be the one that is farther away from the target intersection than the first word position or the second predicted position.

[0263] In some embodiments, predicting relevant locations may include, but is not limited to, starting to predict locations, starting to collect locations, and / or stopping to predict locations.

[0264] After determining the prediction baseline location, the prediction related locations can be determined based on the prediction baseline location.

[0265] For example, the starting prediction position is a location a certain distance above the prediction reference position, meaning the starting prediction position is farther from the target intersection than the prediction reference position. This allows prediction to begin a few seconds earlier. The starting data collection position can be a location a certain distance above the starting prediction position, again to collect driving data a few seconds before the starting prediction position. This ensures sufficient driving data has been collected at the starting prediction position to extract driving features for use in the yaw prediction model. The ending prediction position can be set at a preset distance from the target intersection, such as 200 meters away.

[0266] In an optional implementation of this embodiment, the driving feature further includes historical yaw information of the sample navigation object; the device further includes:

[0267] The first determining module is configured to determine the velocity similarity curves between the velocity curve models in different navigation directions;

[0268] The second determining module is configured to determine whether the target intersection is suitable for predicting yaw behavior based on the speed similarity curve, the displacement curve model, and the historical yaw information.

[0269] In this optional implementation, after obtaining the velocity curve model during model training, the similarity between velocity curve models in different navigation directions can be calculated. If the different velocity curve models are relatively similar, the navigated object before the target intersection may be difficult to distinguish its actual driving direction before the target intersection using the velocity curve model. In this case, the displacement curve model and historical yaw information can be used for further confirmation. If the displacement curve models of the navigated objects in different navigation directions are also relatively similar, and / or the historical yaw rate of the navigated object before the target intersection is low, then the target intersection can be considered unsuitable for yaw behavior prediction, and the target intersection can be marked as an intersection where yaw behavior prediction is not performed. When the navigated object actually uses the navigation service, it does not need to perform yaw behavior prediction when passing through the target intersection. For target intersections that do not meet the above conditions, the target intersection can be marked as suitable for yaw behavior prediction, and the navigated object can perform yaw behavior prediction when passing through the target intersection when actually using the navigation service. It should be noted that the historical deviation information of a sample navigation object can be a comprehensive expression of the historical deviation information of multiple sample navigation objects before the target intersection, such as the average deviation rate of multiple sample navigation objects.

[0270] Figure 8 This is a schematic diagram of the structure of an electronic device suitable for implementing the yaw guidance method and / or yaw prediction model training method according to an embodiment of the present disclosure.

[0271] like Figure 8 As shown, the electronic device 800 includes a processing unit 801, which can be implemented as a CPU, GPU, FPGA, NPU, or other processing unit. The processing unit 801 can execute various processes according to any of the above-described methods of this disclosure, based on a program stored in the read-only memory (ROM) 802 or a program loaded from the storage portion 808 into the random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device 800. The processing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0272] The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 810 as needed so that computer programs read from it can be installed into storage section 808 as needed.

[0273] In particular, according to embodiments of this disclosure, any of the methods described above in the embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program containing program code for performing any of the methods in the embodiments of this disclosure. In such an embodiment, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811.

[0274] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0275] The units or modules described in the embodiments of this disclosure can be implemented in software or hardware. The described units or modules can also be located in a processor, and the names of these units or modules do not necessarily constitute a limitation on the unit or module itself.

[0276] In another aspect, this disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the apparatus described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores one or more programs that are used by one or more processors to perform the methods described in this disclosure.

[0277] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A yaw guidance method, wherein, include: Obtain the driving data of the navigated object before the target intersection; Based on the driving data, the driving characteristics and navigation direction of the navigated object before the target intersection are extracted. The driving characteristics include the correspondence between the speed and distance of the navigated object relative to the target intersection and the lateral offset of the navigated object at different preset distances before the target intersection. The yaw behavior of the navigated object is predicted based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection. The yaw prediction model includes a velocity curve model of speed and distance in different navigation directions and a displacement curve model of lateral offset and distance in different navigation directions, which are pre-fitted and trained for the target intersection. The prediction of the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection specifically includes: Based on the velocity curve model and displacement curve model, the predicted relevant position of the navigated object in front of the target intersection according to the navigation direction is determined. Based on the predicted relevant position, the driving data of the navigated object is collected, and the yaw behavior is predicted. When a deviation is predicted in the navigated object, the navigated object is guided to travel in the correct direction.

2. The method according to claim 1, wherein, The driving data includes the current navigation data of the navigated object before the target intersection and the trajectory data of the navigated object under the navigation of the current navigation data; Based on the driving data, the driving characteristics and navigation direction of the navigated object before the target intersection are extracted, including: Based on the current navigation data, determine the navigation direction of the navigated object before the target intersection; Based on the generated trajectory data, determine the correspondence between the speed of the navigated object and the distance of the navigated object relative to the target intersection, as well as the lateral offset at different preset distances before the target intersection.

3. The method according to claim 1 or 2, wherein, When a veergence is predicted in the navigated object, guiding the navigated object to travel in the correct direction includes: Based on the probability of the navigated object deviating from its course, different guidance methods are used to guide the navigated object to travel in the correct navigation direction; the different guidance methods include one or more combinations of different sound effects, different animations, and different voice playback content.

4. The method according to claim 1 or 2, wherein, Obtain the driving data of the navigated object before the target intersection, including: When the navigated object meets the yaw prediction conditions, the predicted relevant positions before the target intersection included in the yaw prediction model are obtained; Based on the predicted relevant location, the driving data of the navigated object is collected at preset time intervals.

5. A method for training a yaw prediction model, wherein, include: Obtain the driving data of the sample navigation object in front of the target intersection; Based on the driving data, driving characteristics of sample navigation objects traveling in different navigation directions before the target intersection are extracted. The driving characteristics include the speed and lateral deviation of the sample navigation objects at different preset distances during their travel before the target intersection. Based on the driving characteristics in different navigation directions, yaw prediction models are trained in different navigation directions so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection. The yaw prediction models include speed curve models of speed and distance in different navigation directions and displacement curve models of lateral offset and distance in different navigation directions, which are fitted and trained for the target intersection.

6. The method according to claim 5, wherein, Based on the driving data, driving characteristics of sample navigation objects traveling in different navigation directions before the target intersection are extracted, including: Based on the driving data, determine the navigation direction of the sample navigation object before the target intersection; Determine the distribution information of the driving characteristics of the sample navigation object; Based on the distribution information, the driving characteristics of the sample navigation objects in the same navigation direction are determined.

7. The method according to claim 6, wherein, The driving data includes the actual trajectory data generated by the sample navigation object before the target intersection under the navigation of historical navigation data; Determining the distribution information of the driving characteristics of the sample navigation object includes: Based on the actual trajectory data, determine the speed and lateral offset of the sample navigation object when it travels at different preset distances before the target intersection; Based on the speed and lateral offset of multiple sample navigation objects in the same navigation direction, the distribution information of the speed and lateral offset at different preset distances in the same navigation direction is determined.

8. The method according to any one of claims 5-7, wherein, Based on the driving characteristics in different navigation directions, yaw prediction models are trained for different navigation directions so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection, including: Based on the speed fitting of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a speed curve model of speed versus distance in different navigation directions is obtained; Based on the lateral offset of the sample navigation objects in different navigation directions at different preset distances in front of the target intersection, a displacement curve model of lateral offset versus distance in different navigation directions is obtained; Based on the velocity curve model and displacement curve model, the predicted relevant positions before the target intersection in different navigation directions are determined; wherein, when predicting the yaw behavior of the navigated object, the driving data of the navigated object is collected based on the predicted relevant positions, and the yaw behavior is predicted.

9. The method according to claim 8, wherein, Based on the velocity curve model and displacement curve model, the predicted relevant positions before the target intersection in different navigation directions are determined, including: Determine the velocity similarity curves between the velocity curve models in different navigation directions, and determine the displacement similarity curves between the displacement curve models in different navigation directions; The first predicted position is determined based on the inflection point on the velocity similarity curve, and the second predicted position is determined based on the inflection point on the displacement similarity curve; The predicted relevant location is determined based on the first predicted location and the second predicted location.

10. The method according to any one of claims 5-7, wherein, The driving characteristics also include historical yaw information of the sample navigation object; the method further includes: Determine the velocity similarity curves between the velocity curve models described in different navigation directions; Based on the speed similarity curve, the displacement curve model, and the historical yaw information, it is determined whether the target intersection is suitable for predicting yaw behavior.

11. A yaw guidance device, wherein, include: The first acquisition module is configured to acquire the driving data of the navigated object before the target intersection; The first extraction module is configured to extract the driving characteristics and navigation direction of the navigated object before the target intersection based on the driving data. The driving characteristics include the correspondence between the speed and distance of the navigated object relative to the target intersection and the lateral offset of the navigated object at different preset distances before the target intersection. The prediction module is configured to predict the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection. The yaw prediction model includes a velocity curve model for speed versus distance in different navigation directions and a displacement curve model for lateral offset versus distance in different navigation directions, pre-trained and fitted to the target intersection. Specifically, predicting the yaw behavior of the navigated object based on the navigation direction, the driving characteristics, and the yaw prediction model corresponding to the target intersection includes: Based on the velocity curve model and displacement curve model, the predicted relevant position of the navigated object in front of the target intersection according to the navigation direction is determined. Based on the predicted relevant position, the driving data of the navigated object is collected, and the yaw behavior is predicted. The guidance module is configured to guide the navigated object in the correct direction when it is predicted that the navigated object will veer off course.

12. A yaw prediction model training device, wherein, include: The second acquisition module is configured to acquire driving data of the sample navigation object before the target intersection. The second extraction module is configured to extract driving features of sample navigation objects traveling in different navigation directions before the target intersection based on the driving data. The driving features include the speed and lateral offset of the sample navigation objects at different preset distances during their travel before the target intersection. The training module is configured to train yaw prediction models in different navigation directions based on the driving characteristics in different navigation directions, so that the yaw prediction models can predict the yaw behavior of the navigated object before the target intersection. The yaw prediction models include speed curve models of speed and distance in different navigation directions and displacement curve models of lateral offset and distance in different navigation directions, which are fitted and trained for the target intersection.

13. An electronic device, wherein, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method according to any one of claims 1-10.

14. A computer program product comprising computer instructions, wherein, When executed by a processor, the computer instructions implement the method described in any one of claims 1-10.

Images