Behavior prediction method, device, equipment and medium for intersection objects

Abstract

The disclosed embodiments disclose a method, apparatus, device and medium for predicting the behavior of intersection objects. The method comprises: obtaining each intersection departure area corresponding to the intersection to be passed, and for each non-motor vehicle to be predicted object in the intersection to be passed, obtaining the predicted heading probability corresponding to each intersection departure area based on the positional relationship information between the object to be predicted and each intersection departure area, and a pre-constructed regional probability prediction model, and then determining a target area from each intersection departure area based on each predicted heading probability, and determining a target trajectory based on the target area, thereby realizing the prediction of the target area and trajectory of the non-motor vehicle object, and then realizing the behavior prediction of the non-motor vehicle object, solving the problems of lack of behavior prediction of non-motor vehicle obstacles and inability to predict the target area, and solving the problem of poor stability of prediction results using neural networks in the prior art.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent driving technology, and in particular to a method, device, equipment, and medium for predicting the behavior of objects at intersections. Background Art

[0002] In recent years, intelligent driving technology has become a hot topic in the automotive industry. To ensure safe and smooth driving, a crucial yet often overlooked issue is predicting the future behavior of obstacles surrounding the vehicle. Obstacle prediction is a prerequisite for planning and controlling the vehicle's next maneuvers.

[0003] Current obstacle behavior prediction algorithms for motor vehicles (cars, buses, trucks, etc.) directly generate predicted trajectories for a period of time in the future (i.e., from trajectory to trajectory) based on the historical state information of the obstacle. These algorithms primarily use various deep learning algorithms, which require large amounts of data to be processed, labeled, and trained. These algorithms also have weak interpretability and cannot guarantee the stability of the prediction results. Furthermore, they lack semantic information about the intended trajectory, i.e., the target area that the obstacle is about to reach, hindering subsequent planning and control of the vehicle.

[0004] During the implementation of this invention, we discovered that the existing technology suffers from at least the following technical issues: The prediction results are unstable and lack semantic information about intent, hindering subsequent vehicle planning and control. Furthermore, there is a lack of prediction of the behavior of non-motorized vehicle obstacles. Non-motorized vehicles are important traffic participants, and accurate prediction of their intentions and trajectories is crucial for the safe operation of intelligent vehicles. Summary of the Invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a method, device, equipment and medium for predicting the behavior of intersection objects to achieve the prediction of the target area and trajectory of non-motor vehicle objects, thereby achieving the behavior prediction of non-motor vehicle objects, solving the problem that the existing technology lacks behavior prediction of non-motor vehicle obstacles and cannot predict the target area. In addition, by predicting behavior through prediction probability, the problem of poor stability of prediction results using neural networks in the existing technology is solved.

[0006] In a first aspect, an embodiment of the present disclosure provides a method for predicting the behavior of an object at an intersection, characterized in that the method includes:

[0007] Obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0008] For each object to be predicted in the intersection to be passed, obtaining positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

[0009] Inputting each of the position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to each of the intersection departure areas;

[0010] A target area corresponding to the object to be predicted is determined in each of the intersection exit areas based on each of the predicted heading probabilities, and a target trajectory corresponding to the object to be predicted is determined based on the target area.

[0011] In a second aspect, an embodiment of the present disclosure further provides a device for predicting the behavior of an object at an intersection, the device comprising:

[0012] The device for predicting the behavior of an object at an intersection is characterized by comprising:

[0013] An area determination module is used to obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0014] an information acquisition module, configured to acquire, for each object to be predicted in the intersection to be passed, positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

[0015] A probability prediction module is used to input each of the position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to each of the intersection departure areas;

[0016] A behavior determination module is used to determine a target area corresponding to the object to be predicted in each of the intersection exit areas based on each of the predicted going probabilities, and to determine a target trajectory corresponding to the object to be predicted based on the target area.

[0017] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, the one or more processors implement the behavior prediction method for intersection objects as described above.

[0018] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for predicting the behavior of intersection objects.

[0019] The disclosed embodiments provide a method for predicting the behavior of intersection objects. The method obtains each intersection departure area corresponding to the intersection to be passed, and for each non-motor vehicle to be predicted object in the intersection to be passed, obtains the predicted heading probability corresponding to each intersection departure area based on the positional relationship information between the object to be predicted and each intersection departure area, and a pre-constructed regional probability prediction model. Then, a target area is determined from each intersection departure area based on each predicted heading probability, and a target trajectory is determined based on the target area, thereby realizing the prediction of the target area and trajectory of the non-motor vehicle object, and then realizing the behavior prediction of the non-motor vehicle object, solving the problem of the prior art lacking the behavior prediction of non-motor vehicle obstacles and being unable to predict the target area. In addition, the behavior prediction is performed by predicting the probability, solving the problem of the prior art using the neural network prediction result with poor stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0021] Figure 1 This is a flow chart of a method for predicting the behavior of an object at an intersection according to an embodiment of the present disclosure;

[0022] Figure 2 A schematic diagram of exit areas of each intersection corresponding to an intersection to be passed in an embodiment of the present disclosure;

[0023] Figure 3 This is a simple node connection diagram in an embodiment of the present disclosure;

[0024] Figure 4 This is a schematic diagram of node connection incorporating time factors in an embodiment of the present disclosure;

[0025] Figure 5 A schematic diagram of the structure of a dynamic Bayesian model in the embodiment of the present disclosure

[0026] Figure 6 is a schematic diagram of an autonomous driving system according to an embodiment of the present disclosure;

[0027] Figure 7 Schematic diagram of the structure of a device for predicting the behavior of an object at an intersection according to an embodiment of the present disclosure;

[0028] Figure 8 Schematic diagram of the structure of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0030] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0031] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0032] Figure 1 This is a flow chart of a method for predicting the behavior of an object at an intersection according to an embodiment of the present disclosure. This method can be executed by a device for predicting the behavior of an object at an intersection, which can be implemented in software and / or hardware and can be configured in an electronic device. Figure 1 As shown, the method may specifically include the following steps:

[0033] S110: Obtain exit areas of each intersection corresponding to the intersection to be passed by the current vehicle.

[0034] The current vehicle may be an autonomous vehicle that makes decisions based on obstacle behavior. The intersection to be passed may be an intersection that the current vehicle is about to reach. For example, an intersection within the current vehicle's field of view may be considered the intersection to be passed, or an intersection within the current vehicle's field of view and within the current vehicle's route may be considered the intersection to be passed.

[0035] For each intersection to be passed, the corresponding exit areas can be further determined. The exit area can be the area away from the intersection to be passed. The exit area can be defined by a pose (including position and direction information) and a circle centered at the pose point. The size of the circle represents the size of the exit area.

[0036] Specifically, the intersection exit area can be determined based on the lane topology and lane attributes of the intersection. For example, the intersection exit area can be defined centered around the start of the lane leaving the intersection, and the intersection exit area can be defined centered around the ends of the sidewalks surrounding the intersection.

[0037] For example, Figure 2 This is a schematic diagram of the exit areas of each intersection corresponding to a to-be-passed intersection in an embodiment of the present disclosure, see Figure 2 The intersection exit area can be defined as the area at both ends of the sidewalk centerline, or the area at the end of the motorway centerline corresponding to the direction of travel. The definition of the intersection exit area can be the same for different predicted objects at the same intersection.

[0038] It should be noted that in this embodiment, the intersection exit areas corresponding to different intersections to be passed may not be exactly the same. For example, corresponding intersection exit areas can be pre-set for each intersection type. Then, when determining the intersection exit areas corresponding to the intersection to be passed, the intersection type corresponding to the intersection to be passed is first determined, and then the intersection exit areas corresponding to the intersection type are matched.

[0039] S120. For each object to be predicted in the intersection to be passed, obtain positional relationship information between the object to be predicted and the exit area of ​​each intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed.

[0040] The object to be predicted may be a non-motor vehicle object located in the intersection to be passed, that is, a non-motor vehicle type obstacle of the current vehicle, such as a pedestrian, bicycle or tricycle.

[0041] Specifically, all non-motor vehicle-type objects within the intersection to be passed can be used as objects to be predicted, so that their behaviors can be predicted, thereby facilitating the current vehicle to make autonomous driving decisions based on the predicted behaviors.

[0042] In this embodiment, for each object to be predicted, it is necessary to obtain positional relationship information between the object to be predicted and each intersection exit area. This positional relationship information can describe the relationship between the location of the object to be predicted and the location of the intersection exit area. For example, the positional relationship information can include distance, angle difference, whether the object to be predicted is located in front of the intersection exit area, etc.

[0043] Optionally, the positional relationship information between the object to be predicted and each intersection departure area is obtained, including: for each intersection departure area, obtaining at least one of the current distance between the object to be predicted and the intersection departure area, the current angle difference between the object to be predicted and the intersection departure area, and the historical average distance from the object to be predicted to the lane line corresponding to the intersection departure area; and using at least one of the obtained current distance, current angle difference and historical average distance as the positional relationship information between the object to be predicted and the intersection departure area.

[0044] The current distance can be the Euclidean distance between the object to be predicted and the intersection exit area. The current angle difference can be the angle difference between the orientation of the object to be predicted and the intersection exit area. The historical average distance can be the average distance from the historical driving trajectory of the object to be predicted to the lane line corresponding to the intersection exit area.

[0045] Specifically, at least one of the current distance, the current angle difference, and the historical average distance can be obtained, and the positional relationship information can be determined based on the obtained information. Through the above method, accurate determination of the positional relationship information is achieved. Using Euclidean distance, angle difference, or historical average distance as the positional relationship information can further improve prediction accuracy.

[0046] In this embodiment, in addition to using Euclidean distance, angle difference or historical average distance as position relationship information, other information may also be selected as position relationship information, which is not limited in this embodiment.

[0047] Optionally, obtaining the positional relationship information between the object to be predicted and each intersection departure area also includes: for each intersection departure area, determining whether the intersection departure area is located behind the object to be predicted, and if so, determining that the first variable corresponding to the intersection departure area is a true value; determining whether the object to be predicted is located within a preset area corresponding to the intersection departure area, and if so, determining that the second variable corresponding to the intersection departure area is a true value, wherein the preset area is a related area of ​​the lane connected to the intersection departure area; determining the first variable and the second variable corresponding to the intersection departure area as the positional relationship information between the object to be predicted and the intersection departure area.

[0048] Among them, the preset area can be an adjacent area of ​​the lane connected to the intersection exit area. For example, the lane connected to the intersection exit area can be taken as the center, and the preset area corresponding to the intersection exit area can be determined according to the size of the preset area.

[0049] Specifically, the first and second variables can be Boolean variables, i.e., binary scalars that take on the values ​​of true or false. If the intersection exit area is located behind the object to be predicted, the first variable takes on the value true. If the object to be predicted is within the preset area corresponding to the intersection exit area, indicating that the object to be predicted is near a lane connected to the intersection exit area, the second variable takes on the value true.

[0050] Furthermore, the first and second true values ​​can be determined as positional relationship information between the object to be predicted and the intersection exit area. This approach also includes determining whether the intersection exit area is behind the object to be predicted and whether the object to be predicted is near a lane connected to the intersection exit area, enriching the positional relationship information and further improving prediction accuracy.

[0051] S130: Input each position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to the exit area of ​​each intersection.

[0052] In this embodiment, the object to be predicted located at the intersection to be passed needs to move from its current location to a certain intersection exit area in order to leave the intersection. Therefore, this embodiment can use a regional probability prediction model to predict the probability of the object to be predicted leaving the intersection to be passed from each intersection exit area. In other words, the probability of the object to be predicted moving from its current location to each intersection exit area is predicted, and the predicted moving probability corresponding to each intersection exit area is obtained.

[0053] Among them, the regional probability prediction model can be a probability graph model, such as BN (Baysian Network), DBN (Dynamic Bayesian Network), ridge regression model, least squares method and other models.

[0054] Different from the prediction based on the weights and biases of neurons in the neural network, in this embodiment, the regional probability prediction model can obtain the prior probability and conditional probability based on the input position relationship information, and then determine the posterior probability based on the prior probability and conditional probability to obtain the predicted heading probability corresponding to the exit area of ​​each intersection.

[0055] In a specific embodiment, the regional probability prediction model is a dynamic Bayesian model. The position relationship information is input into a pre-built regional probability prediction model to obtain the predicted probability of leaving the area at each intersection. The prediction model may include:

[0056] For each intersection departure area, the intersection departure area is used as the intention node in the dynamic Bayesian model, and at least one of the current distance, current angle difference and historical average distance corresponding to the intersection departure area is used as the observation argument nodes connected to the intention node; for each intention node in the dynamic Bayesian model, the predicted heading probability corresponding to the intention node is determined based on the observation argument nodes connected to the intention node.

[0057] The dynamic Bayesian model can include intention nodes and observation evidence nodes; each intention node represents an intersection departure area, and an observation evidence node represents the current distance, current angle difference, or historical average distance. For example, if the position relationship information includes the current distance, current angle difference, and historical average distance, then each intention node (intersection departure area) is connected to three observation evidence nodes (the current distance, current angle difference, and historical average distance corresponding to the intersection departure area).

[0058] It should be noted that each intention node is connected to each observation argument node, which means that there is a causal relationship between the intention node and the observation argument node. Changes in the intention node will lead to changes in the observation argument node. For example, Figure 3 This is a simple node connection diagram in an embodiment of the present disclosure, wherein the circle in the figure is called a node, m_i represents an intention node, e represents an observation argument node, and the arrows in the figure represent the causal relationship between the nodes.

[0059] In this embodiment, the predicted going probability corresponding to each intention node, that is, the predicted going probability corresponding to each intersection exit area, can be determined based on each observation evidence node connected to each intention node.

[0060] In other words, this embodiment achieves: when e is observed, the intention node m=m is calculated. i The probability size P(m=m i |e), this probability is the predicted going probability corresponding to the intent node. This approach enables the prediction of the predicted going probability corresponding to each intersection's exit area based on a dynamic Bayesian model, thereby enabling the prediction of the intention of each target. Compared to neural network prediction methods, this approach requires no data processing or annotation, and provides stable prediction results.

[0061] Optionally, for each intention node in the dynamic Bayesian model, the predicted going probability corresponding to the intention node is determined based on the observation argument nodes connected to the intention node, which can be: for each intention node, the first conditional probability of each observation argument node under the intention node, the prior probability of the intention node and the marginal probability of each observation argument node connected to the intention node are determined; the product of each first conditional probability and the prior probability is determined, and the ratio of the product to the marginal probability is used as the predicted going probability corresponding to the intention node.

[0062] Among them, for each observation argument node connected to each intention node, the observation argument conditional probability under the intention can be determined to obtain the first conditional probability, such as P(e|m=m i ), indicating that at the intention node m i The first conditional probability of the observation argument node under . The prior probability of each intention node can be pre-set. The marginal probability of each observation argument node connected to the intention node can be understood as a normalized variable.

[0063] For example, the predicted going probability corresponding to the intent node can be calculated by the following formula:

[0064]

[0065] Where P(m=m i |e) indicates the intention node m iThe corresponding predicted probability is the posterior probability; P(e|m=m i ) is the intention node m i The first conditional probability of the observation argument node under the condition, if the number of observation argument nodes connected to the intention node is multiple, then all the first conditional probabilities are multiplied; P(m=m i ) represents the intention node m i Prior probability of ∑ m P(m,e) represents the edge probability of each observation argument node connected to the intention node.

[0066] Through the above method, the predicted going probability corresponding to each intention node can be calculated, and the probability of the predicted object going to the exit area of ​​each intersection can be accurately predicted.

[0067] It should be noted that, in addition to directly obtaining the prior probability of the above-mentioned intention node, the prior probability of the intention node can be further determined by combining the changes brought about by the time factor.

[0068] Optionally, determining the prior probability of the intention node includes: obtaining the predicted going probability corresponding to each intention node at the previous moment, and the conversion probability of each intention node at the previous moment to the intention node at the current moment; for each intention node at the previous moment, taking the product of the predicted going probability corresponding to the intention node at the previous moment and the conversion probability as the reference probability of the intention node at the previous moment, and determining the sum of the reference probabilities of each intention node at the previous moment as the prior probability of the intention node at the current moment.

[0069] That is, the prior probability of the intention node at the current moment can be determined by combining the predicted going probabilities corresponding to each intention node at the previous moment and the conversion probabilities of each intention node at the previous moment to the intention node at the current moment.

[0070] For example, Figure 4 This is a schematic diagram of node connection incorporating time factors in an embodiment of the present disclosure, see Figure 4 , the calculation of the intention probability at the current moment t is affected not only by the observation evidence at the current moment t, but also by the calculation result at the previous moment t-1 (i.e. Figure 4 dashed line in the middle).

[0071] Specifically, for each intention node at the previous moment, the product of its corresponding predicted going probability and conversion probability is taken as the reference probability, and then the sum of the reference probabilities of all intention nodes at the previous moment is determined as the prior probability of the intention node at the current moment.

[0072] For example, the following formula can be used to calculate the prior probability of the intention node at the current moment:

[0073] P(m t =m i )=∑ j (P(m t =m i |m t-1 =m j )P(m t-1 =m j |e t-1 ));

[0074] Among them, P(m t =m i ) represents the intention node m at the current time t i The prior probability, P(m t =m i |m t-1 =m j ) represents the intention node m at the previous time t-1 j Transition to the intention node m at the current time t i The transition probability, P(m t-1 =m j |e t-1 ) represents the intention node m at the previous time t-1 j The corresponding predicted probability of going.

[0075] From the calculation formula of the prior probability above, we can know that the predicted probability of going to the intention node at the current moment can be obtained by the following formula:

[0076]

[0077] Among them, P(m t =m i |e t ) represents the intention node m at the current moment i The corresponding predicted probability of going, e t is the intention node m i The corresponding observation argument node, P(e t |m t =m i ) represents the first conditional probability.

[0078] In the above embodiment, the prior probability of the current moment is determined based on the prediction result of the previous moment, thereby considering the changes brought about by the time factor in the intention prediction, and further improving the accuracy of the intention prediction.

[0079] In addition to determining the prior probability of the current moment by combining the prediction results of the previous moment as described above, a causal argument node can be further defined in the dynamic Bayesian model to consider the impact of the causal argument node on the prior probability of the current moment.

[0080] Optionally, after treating each intersection departure area as an intention node in the dynamic Bayesian model, the method further includes: treating the first variable and the second variable corresponding to the intersection departure area as cause argument nodes connecting the intention node;

[0081] Correspondingly, determining the prior probability of the intention node also includes: determining the second conditional probabilities of the intention node under each cause argument node connecting the intention node; taking the sum of the second conditional probabilities as the first value, taking the sum of the reference probabilities of the intention nodes at the previous moment as the second value, and taking the product of the first value and the second value as the prior probability of the intention node at the current moment.

[0082] Among them, the first variable and the second variable corresponding to the intersection departure area can serve as the cause argument nodes corresponding to the intersection departure area, and are connected to the intention node corresponding to the intersection departure area, forming a causal relationship with the intention node, and the change of the cause argument node leads to the change of the intention node.

[0083] For example, Figure 5 This is a structural diagram of a dynamic Bayesian model in an embodiment of the present disclosure, wherein REV and IN are cause argument nodes, representing the first variable and the second variable respectively, a_1, a_2, ... a_m represent each intention node, d_to_area, theta_diff, avg_d_to_lane are observation argument nodes, representing the current distance, the current angle difference and the historical average distance respectively; the dotted line connecting each intention node represents the prediction result at the previous moment.

[0084] Specifically, the prior probability of the intention node at the current moment can be determined by combining the second conditional probabilities of the intention nodes and the reference probabilities of the intention nodes at the previous moment. The second conditional probability represents the conditional probability of the intention under specific cause arguments.

[0085] In this embodiment, the sum of the two second conditional probabilities is taken as the first value; the sum of the reference probabilities of all intention nodes at the previous moment is taken as the second value; the product of the first value and the second value is the prior probability of the intention node at the current moment.

[0086] For example, the prior probability can be calculated by the following formula:

[0087]

[0088] Among them, P(m t=a1) represents the prior probability of the intention node a1 at the current moment, and c represents the two causal argument nodes that affect the intention node: the first variable and the second variable. Since the objective results of the first variable and the second variable can be obtained before calculation, their prior probabilities P(c) can be regarded as 1. Finally, the prior probability of the intention node at the current moment can be expressed by the following formula:

[0089]

[0090] Furthermore, the predicted probability of going to the intention node at the current moment can be calculated by the following formula:

[0091]

[0092] in, Observational evidence under specific intention a1 Conditional probability, that is The first conditional probability, P(m t =a1|m t-1 =m j ) represents the intention node m at the previous time t-1 j The probability of transitioning to the intention node a1 at the current time t, P(m t =a1|c) represents the conditional probability of intention under the specific causal argument c, that is, the second conditional probability of a1. Represents the edge probability of each observation argument node connected to the intention node a1.

[0093] In the above embodiment, by introducing the causal argument node into the dynamic Bayesian model to consider the impact of the causal argument node on the prior probability at the current moment, the accuracy of the prior probability is further improved, thereby improving the accuracy of the prediction result at the current moment.

[0094] It should be noted that the above-mentioned first conditional probability, second conditional probability and transition probability can be set manually or obtained through training.

[0095] In a specific embodiment, before inputting each position relationship information into a pre-built regional probability prediction model to obtain the predicted going probability corresponding to each intersection leaving area, it also includes: obtaining a training data set, wherein the training data set includes sample position relationship information between the sample object and each sample leaving area, and each area going probability label corresponding to each sample position relationship information; training the regional probability prediction model based on the training data set to obtain a first conditional probability table, a second conditional probability table and a conversion probability table, or obtaining a first conditional probability model, a second conditional probability model and a conversion probability model.

[0096] The first conditional probability table or the first conditional probability model is used to determine each first conditional probability, the second conditional probability table or the second conditional probability model is used to determine each second conditional probability, and the conversion probability table or the conversion probability model is used to determine each conversion probability.

[0097] Specifically, the training data set may be a small sample data set, including sample position relationship information corresponding to the region where each sample leaves, and a probability label of going to each region corresponding to the position relationship information of each sample.

[0098] The training data set can be input into the regional probability prediction model so that the regional probability prediction model can obtain the first conditional probability table, the second conditional probability table and the conversion probability table according to the probability labels of each region, or obtain the first conditional probability model, the second conditional probability model and the conversion probability model.

[0099] After obtaining the above probability table or probability model, you can use the first conditional probability table or first conditional probability model to query the first conditional probability of an observation argument node under a certain intention node. You can use the second conditional probability table or second conditional probability model to query the second conditional probability of an intention node under a certain cause argument node. You can also use the transition probability table or transition probability model to query the transition probability between intention nodes. Based on the above queried probabilities, you can further calculate the marginal probabilities and then the posterior probabilities to obtain the predicted heading probability corresponding to the intersection departure area.

[0100] By training the training data set in the above manner, the first conditional probability, the second conditional probability and the transition probability can be accurately determined without manual setting, which improves the prediction efficiency while ensuring the accuracy of each probability.

[0101] It should be noted that the above method of calculating the predicted probability of travel based on the first conditional probability, the second conditional probability, and the conversion probability can be understood as: based on the known probability of various observational evidence (position, angle difference, etc.) of traveling to each specific area, predicting the probability of traveling to each area under known observational evidence.

[0102] S140 : determining a target area corresponding to the object to be predicted in each intersection exit area based on each predicted going probability, and determining a target trajectory corresponding to the object to be predicted based on the target area.

[0103] Specifically, after obtaining the predicted probability of the object to be predicted heading to each intersection exit area, a target area can be determined from each intersection exit area based on all the predicted probability of heading.

[0104] For example, all predicted going probabilities can be sorted in descending order of probability, and the intersection exit areas corresponding to the top N predicted going probabilities can be used as the target area, or the intersection exit area corresponding to the largest predicted going probability can be used as the target area.

[0105] After obtaining the target area, the target trajectory of the object to be predicted towards the target area can be predicted for each target area. For example, the target trajectory can be generated using a curve model or a deep learning network. To further improve the reliability of the target trajectory, this embodiment can also combine environmental information to predict the target trajectory.

[0106] In a specific embodiment, determining the target trajectory corresponding to the object to be predicted based on the target area may include the following steps:

[0107] Step 11: Acquire state information and environmental information of the object to be predicted at the current moment, wherein the state information includes first position information and first speed information of the object to be predicted, and the environmental information includes second position information and second speed information of other objects to be predicted at the intersection to be passed;

[0108] Step 12: Determine the speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment, and determine the state information of the object to be predicted at the next moment based on the state information and speed control information of the object to be predicted at the current moment;

[0109] Step 13: Take the next moment as the current moment, reacquire the environmental information at the current moment, and determine the speed control information based on the environmental information and state information at the current moment, until the state information of the object to be predicted at the last moment is determined, and determine the target trajectory corresponding to the object to be predicted based on the state information at each moment; wherein, the first position information of the object to be predicted at the last moment is the target area.

[0110] For the above step 11, considering that there are multiple objects to be predicted in the intersection to be passed, the positions and speeds of each object to be predicted will affect each other. Therefore, when predicting the target trajectory of the current object to be predicted, not only the state information of the object to be predicted at the current moment, that is, the first position information and first speed information of the object to be predicted, but also the environmental information of the object to be predicted at the current moment, that is, the second position information and second speed information of other objects to be predicted, can be obtained.

[0111] In step 12 above, based on the current environmental information and state information, the optimal control variable for the object to be predicted at the current moment, i.e., the speed control information, can be determined. Furthermore, the state information at the next moment is determined based on the current speed control information. The speed control information can be in the form of a speed vector.

[0112] Furthermore, the next moment is used as the current moment, and the environmental information at the current moment is reacquired. The process then returns to step 12 until the state information of the object to be predicted at the final moment is predicted. The first position information at the final moment corresponds to the target area, indicating that the object to be predicted has reached the target area at the final moment. Ultimately, the target trajectory can be determined based on the state information at all moments.

[0113] Through steps 11-13 above, the state information at each moment is predicted, and the environmental information at each moment is taken into account, ensuring the accuracy of the trajectory prediction. In addition, compared with deep learning or curve model methods, after determining the state information at each moment, the above steps can update the environmental information, and then solve the state information at the next moment based on the updated environmental information, and so on. Therefore, the above method can take into account the dynamic changes of the environment, generate more reasonable predicted trajectories, and further improve the accuracy of the trajectory.

[0114] With respect to the above step 12, optionally, the speed control information of the object to be predicted at the current moment is determined based on the environmental information and state information at the current moment, including: constructing a distance cost function, wherein the distance cost function is used to describe the relationship between the object distance and cost between the object to be predicted and other objects to be predicted; with the goal of minimizing the calculation result of the distance cost function, the speed control information corresponding to the minimum distance cost is determined based on the environmental information and state information at the current moment.

[0115] Specifically, a distance cost function can be used to calculate a distance cost based on the distance between the object to be predicted and other objects to be predicted. The purpose of constructing a distance cost function is to consider the interaction between the object to be predicted and other objects in the environment. To avoid collisions between the object to be predicted and other objects to be predicted, the minimum distance between the object to be predicted and other objects to be predicted should be as large as possible, thereby minimizing the distance cost.

[0116] In this embodiment, the goal is to minimize the calculated result of the distance cost function, that is, to minimize the distance cost, and obtain speed control information corresponding to the minimum distance cost. By minimizing the calculated result of the distance cost function as the goal, the optimal speed control information is obtained, so that the speed control information can ensure that the distance between objects is as large as possible, thereby avoiding collisions between objects.

[0117] Regarding the process of constructing a distance cost function, in a specific implementation, constructing the distance cost function includes: constructing a relationship between the minimum object distance between the object to be predicted and other objects to be predicted and speed control information based on the change in the square distance between the object to be predicted and other objects to be predicted; constructing a distance cost function based on the relationship between the minimum object distance and the speed control information and a preset first hyperparameter.

[0118] The squared distance change is used to describe the distance change between objects at each moment. In this embodiment, assuming that the other objects to be predicted maintain their current uniform speed, the squared distance change between the object to be predicted and the other objects to be predicted can be expressed as follows:

[0119] d mn (t,u m )=‖p m +tu m -p n -v n ‖ 2 ;

[0120] Among them, t represents the current time, u m Indicates speed control information, p m Represents the object to be predicted o m The first position information, p n Represents other objects to be predicted o n The second position information, v n Represents other objects to be predicted o n The second speed information.

[0121] Furthermore, the formula for the change in squared distance can be derived with respect to time t to obtain the minimum distance between the object to be predicted and other objects to be predicted that will occur at time:

[0122]

[0123] Furthermore, by substituting the formula obtained after taking the derivative of time t into the formula for square distance change, we can obtain the relationship between the minimum object distance and speed control information:

[0124]

[0125] Where k = p m -p n ,q=u m -v n Furthermore, a distance cost function can be constructed based on the preset first hyperparameter and the relationship. The distance cost function can be expressed as follows:

[0126]

[0127] Among them, σ d To preset the first hyperparameter, control other objects to be predicted o n The influence range of σ d When the larger the value is, the more m Tends to be further away from other objects to be predicted n .

[0128] Specifically, the distance cost function can be solved by the gradient descent method to obtain the speed control information. Through the above method, the distance cost function is accurately constructed, thereby ensuring the accuracy of the speed control information solved based on the distance cost function.

[0129] Considering that there may be multiple other objects to be predicted in the environment, and multiple other objects to be predicted have an impact on the object to be predicted, a final distance cost function can also be constructed. The final distance cost function is the weighted sum of the costs of each other object to be predicted.

[0130] Optionally, after constructing the distance cost function based on the relationship between the minimum object distance and the speed control information and the preset first hyperparameter, it also includes: if there are multiple other objects to be predicted, for each other object to be predicted, according to the angle difference and distance between the object to be predicted and the other objects to be predicted at the current moment, determine the cost weight corresponding to the other objects to be predicted, and determine the distance cost components corresponding to the other objects to be predicted based on the distance cost function and the cost weights corresponding to the other objects to be predicted; and construct the final distance cost function based on the sum of the distance cost components corresponding to each other object to be predicted.

[0131] The angle difference may be the angle that the current direction of the object to be predicted rotates to the other object to be predicted, and the distance may be the distance between the two objects at the current moment.

[0132] For example, the cost weights corresponding to other objects to be predicted can be calculated according to the following formula:

[0133]

[0134] Among them, w mn Represents other objects to be predicted o nThe corresponding cost weight, φ is the object o to be predicted at the current moment m With other objects to be predicted o n The angle difference between them, the definition of k follows the above description, β, σ w is a hyperparameter, β determines the maximum value of the distance cost term, σ w Action and σ d Similarly, it affects the scope of other objects to be predicted.

[0135] After obtaining the cost weights corresponding to each other object to be predicted, the product of the cost weights corresponding to the other objects to be predicted and the distance cost function can be used as the distance cost component corresponding to the other object to be predicted, and then the sum of the distance cost components corresponding to all other objects to be predicted can be used as the final distance cost function.

[0136] For example, at t i At time, the object to be predicted o m The distance cost function to the environment is defined as:

[0137] I m (u m )=∑ n≠m w mn ·E mn (u m );

[0138] Among them, I m (u m ) is the final distance cost function. Specifically, the final distance cost function can be solved using the gradient descent method to obtain speed control information. Through the above method, the distance cost function for multiple other objects to be predicted is constructed, thereby achieving trajectory prediction that takes into account whether the predicted object will avoid colliding with other objects, further improving the accuracy of trajectory prediction.

[0139] In addition to solving the optimal control amount according to the distance cost function, in this embodiment, the optimal control amount can also be solved in combination with the speed cost function.

[0140] For example, optionally, determining the speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment also includes: constructing a speed cost function, wherein the speed cost function is used to describe the relationship between the speed difference and cost between the speed control information of the object to be predicted and the expected speed; constructing a total cost function based on the speed cost function and the distance cost function, with the goal of minimizing the calculation result of the total cost function, and determining the speed control information corresponding to the minimum total cost based on the environmental information and state information at the current moment.

[0141] Specifically, the speed cost function can be used to calculate the corresponding speed cost based on the speed difference between the speed control information and the expected speed of the object to be predicted. The purpose of constructing the speed cost function is to make the speed control information as close to the expected speed as possible. The smaller the speed difference between the speed control information and the expected speed, the smaller the speed cost calculated by the speed cost function.

[0142] For example, the constructed speed cost function can be expressed as follows:

[0143]

[0144] Among them, S m (u m ) represents the speed cost, Indicates the expected speed. You can pre-set the expected speed for each object type, and then match the expected speed of the object to be predicted.

[0145] Furthermore, the sum of the speed cost function and the distance cost function can be used as the total cost function; or, the total cost function can be constructed based on the speed cost function, its corresponding weight, the distance cost function, and its corresponding weight. Specifically, the total cost function can be solved using the gradient descent method to obtain the speed control information.

[0146] Furthermore, with the goal of minimizing the total cost calculated by the total cost function, the optimal control variable is solved to obtain speed control information. This approach achieves a combined control variable solution for speed and distance costs, taking into account that the speed of the object to be predicted should theoretically be close to the expected speed at each moment and theoretically avoid collisions with other objects, further improving the accuracy of trajectory prediction.

[0147] It should be noted that, in this embodiment, in addition to combining the speed cost function, the direction cost function may also be combined.

[0148] For example, optionally, before constructing the total cost function based on the speed cost function and the distance cost function, it also includes: constructing a direction cost function, wherein the direction cost function is used to describe the relationship between the direction difference and the cost between the speed control information of the object to be predicted and the target area; accordingly, constructing the total cost function based on the speed cost function and the distance cost function includes: constructing the total cost function based on the speed cost function, the direction cost function and the distance cost function.

[0149] The directional cost function can be used to calculate the corresponding directional cost based on the directional difference between the speed control information and the target area. The purpose of constructing the directional cost function is to make the speed direction in the speed control information as close as possible to the direction of the target area. The smaller the directional difference between the speed control information and the target area, the smaller the directional cost calculated by the directional cost function.

[0150] For example, the constructed direction cost function can be expressed as follows:

[0151]

[0152] Among them, z m Represents the coordinate point in the target area, for example, it can be the edge point of the area closest to the object to be predicted in the target area, or it can be the center point of the area of ​​the target area. m (u m ) represents the direction cost.

[0153] Furthermore, a total cost function can be constructed based on the speed cost function, the direction cost function, and the distance cost function. Specifically, the sum of the speed cost function, the direction cost function, and the distance cost function can be directly determined as the total cost function; or, the total cost function can be constructed based on the speed cost function, the weight corresponding to the speed cost function, the direction cost function, the weight corresponding to the direction cost function, and the distance cost function.

[0154] For example, the total cost function can be expressed as follows:

[0155] E m (u m )=I m (u m )+λ1S m (u m )+λ2D m (u m );

[0156] Among them, E m (u m ) represents the total cost, and λ1 and λ2 are hyperparameters used to adjust the impact of the speed and direction costs. Specifically, this formula can be solved using gradient descent to obtain speed control information. This approach allows for a solution that combines the speed, direction, and distance costs. This approach takes into account that the speed of the object being predicted at each moment should theoretically be close to the desired speed, avoid collisions with other objects, and approach the direction of the target area, further improving trajectory prediction accuracy.

[0157] In this embodiment, the speed control information at each moment is obtained, and then the state information at the next moment is obtained based on the speed control information at the moment. Similarly, the state information at a total of N moments can be obtained.

[0158] For the above step 13, in a specific implementation, based on the state information and speed control information of the object to be predicted at the current moment, the state information of the object to be predicted at the next moment is determined, including: determining the first speed information of the object to be predicted at the next moment according to the preset second hyperparameter, the first speed information of the object to be predicted at the current moment, and the speed control information; determining the first position information of the object to be predicted at the next moment according to the first position information of the object to be predicted at the current moment, the first speed information at the next moment, and the time difference between the current moment and the next moment.

[0159] Specifically, the product of the first speed information and a preset second hyperparameter can be determined, and the product of the result of subtracting the set value from the preset second hyperparameter and the speed control information can be determined, and the sum of the two products is used as the first speed information at the next moment. The set value can be 1.

[0160] Furthermore, the product of the first speed information at the next moment and the time difference may be determined, and the sum of the product and the first position information at the current moment may be used as the first position information at the next moment.

[0161] For example, see the following formula:

[0162]

[0163] Among them, the state information of the object to be predicted at the current moment is defined as Indicates the first position information at the current moment, which can be a two-dimensional coordinate; The first speed information at the current moment may be a speed vector. Indicates the first position information at the next moment, represents the first speed information at the next moment, α represents the preset second hyperparameter, and its value is between 0 and 1.

[0164] In the above embodiment, based on the speed control information at the current moment, the state information at the next moment is determined. By determining the state information at the next moment based on the state information and speed control information at the previous moment, the state information at each moment is determined in sequence, thereby ensuring the correlation between the state information at each moment and further improving the accuracy of trajectory prediction.

[0165] It should be noted that in the formulas provided in this embodiment, the hyperparameters α, β, σ involved are d ,σ w ,λ1,λ2, a large amount of trajectory data of non-motor vehicles in the intersection area can be collected in advance, and the various hyperparameters can be estimated by the maximum likelihood method.

[0166] Through the method provided in this embodiment, the target area of ​​each object to be predicted and the target trajectory to each target area can be predicted, thereby realizing the behavior prediction of each object to be predicted. When an autonomous driving vehicle passes through an intersection, the method provided in this embodiment can be used to automatically predict the intention and trajectory of non-motor vehicles, so that the autonomous driving vehicle can make more reasonable decisions and plans, and achieve a safer and smoother driving effect.

[0167] For example, the method provided in this embodiment can be executed by a behavior prediction module of an autonomous driving vehicle. Figure 6 , Figure 6 This is a schematic diagram of an autonomous driving system in an embodiment of the present disclosure, in which the behavior prediction module can perform intention prediction and trajectory prediction for each object to be predicted based on the information output by the perception module, map module, and positioning module. The planning decision module then makes reasonable path decision planning based on the prediction results, and finally controls the vehicle movement through the control module.

[0168] The behavior prediction method for intersection objects provided in this embodiment obtains each intersection departure area corresponding to the intersection to be passed, and for each non-motor vehicle to be predicted object in the intersection to be passed, obtains the predicted heading probability corresponding to each intersection departure area based on the positional relationship information between the to-be-predicted object and each intersection departure area, and a pre-constructed regional probability prediction model, and then determines the target area from each intersection departure area based on each predicted heading probability, and determines the target trajectory based on the target area, thereby realizing the prediction of the target area and trajectory of the non-motor vehicle object, and then realizing the behavior prediction of the non-motor vehicle object, solving the problem of the prior art lacking behavior prediction of non-motor vehicle obstacles and being unable to predict the target area, and, by performing behavior prediction by means of predicted probability, solving the problem of poor stability of prediction results using neural networks in the prior art.

[0169] Figure 7 FIG. 1 is a schematic diagram of a device for predicting the behavior of an intersection object in an embodiment of the present disclosure. Figure 7 As shown: the device includes: an area determination module 710, an information acquisition module 720, a probability prediction module 730 and a behavior determination module 740.

[0170] The area determination module 710 is used to obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0171] An information acquisition module 720 is configured to acquire, for each object to be predicted at the intersection to be passed, positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle located at the intersection to be passed;

[0172] The probability prediction module 730 is used to input the position relationship information into a pre-built regional probability prediction model to obtain the predicted heading probability corresponding to the exit area of ​​each intersection;

[0173] The behavior determination module 740 is configured to determine a target area corresponding to the object to be predicted in each of the intersection exit areas based on each of the predicted going probabilities, and determine a target trajectory corresponding to the object to be predicted based on the target area.

[0174] The device for predicting the behavior of an object at an intersection provided by the embodiment of the present disclosure can execute the steps of the method for predicting the behavior of an object at an intersection provided by the embodiment of the method of the present disclosure, and the execution steps and beneficial effects are not repeated here.

[0175] Figure 8 This is a schematic diagram of the structure of an electronic device in the embodiment of the present disclosure. Figure 8 , which shows a structural diagram of an electronic device 500 suitable for implementing the embodiments of the present disclosure. Figure 8 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0176] like Figure 8 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes to implement the methods of the embodiments described in the present disclosure according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 into the random access memory (RAM) 503. Various programs and data required for the operation of the electronic device 500 are also stored in the RAM 503. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0177] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart, thereby implementing the behavior prediction method of the intersection object as described above. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0178] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0179] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device:

[0180] Obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0181] For each object to be predicted in the intersection to be passed, obtaining positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

[0182] Inputting each of the position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to each of the intersection departure areas;

[0183] A target area corresponding to the object to be predicted is determined in each of the intersection exit areas based on each of the predicted heading probabilities, and a target trajectory corresponding to the object to be predicted is determined based on the target area.

[0184] Optionally, when the above one or more programs are executed by the electronic device, the electronic device may also execute other steps described in the above embodiments.

[0185] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0186] Solution 1: A method for predicting the behavior of objects at an intersection, the method comprising:

[0187] Obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0188] For each object to be predicted in the intersection to be passed, obtaining positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

[0189] Inputting each of the position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to each of the intersection departure areas;

[0190] A target area corresponding to the object to be predicted is determined in each of the intersection exit areas based on each of the predicted heading probabilities, and a target trajectory corresponding to the object to be predicted is determined based on the target area.

[0191] Solution 2: According to the method of Solution 1, obtaining the positional relationship information between the object to be predicted and each of the intersection exit areas includes:

[0192] For each intersection exit area, obtaining at least one of a current distance between the object to be predicted and the intersection exit area, a current angle difference between the object to be predicted and the intersection exit area, and a historical average distance from the object to be predicted to a lane line corresponding to the intersection exit area;

[0193] At least one of the acquired current distance, the current angle difference, and the historical average distance is used as positional relationship information between the object to be predicted and the intersection exit area.

[0194] Solution 3: According to the method of Solution 2, the regional probability prediction model is a dynamic Bayesian model, and the inputting of each position relationship information into a pre-built regional probability prediction model to obtain the predicted heading probability corresponding to each intersection departure area includes:

[0195] For each intersection departure area, the intersection departure area is used as an intention node in the dynamic Bayesian model, and at least one of the current distance, the current angle difference, and the historical average distance corresponding to the intersection departure area is used as each observation argument node connected to the intention node;

[0196] For each of the intention nodes in the dynamic Bayesian model, the predicted going probability corresponding to the intention node is determined based on the observation argument nodes connected to the intention node.

[0197] Solution 4: According to the method of Solution 3, for each intention node in the dynamic Bayesian model, determining the predicted going probability corresponding to the intention node based on each of the observation argument nodes connected to the intention node includes:

[0198] For each of the intention nodes, determine the first conditional probability of each of the observation argument nodes under the intention node, the prior probability of the intention node, and the marginal probability of each of the observation argument nodes connected to the intention node;

[0199] Determine the product of each of the first conditional probabilities and the prior probability, and use the ratio of the product to the edge probability as the predicted going probability corresponding to the intention node.

[0200] Solution 5: According to the method described in Solution 4, determining the prior probability of the intention node includes:

[0201] Obtain the predicted going probability corresponding to each intention node at the previous moment, and the conversion probability of each intention node at the previous moment to the intention node at the current moment;

[0202] For each intention node at the previous moment, the product of the predicted going probability and the conversion probability corresponding to the intention node at the previous moment is taken as the reference probability of the intention node at the previous moment, and the sum of the reference probabilities of each intention node at the previous moment is determined as the prior probability of the intention node at the current moment.

[0203] Solution 6: According to the method of Solution 5, obtaining the positional relationship information between the object to be predicted and each of the intersection exit areas further includes:

[0204] For each of the intersection exit areas, determining whether the intersection exit area is located behind the object to be predicted, and if so, determining that a first variable corresponding to the intersection exit area is a true value;

[0205] Determining whether the object to be predicted is located within a preset area corresponding to the intersection exit area, and if so, determining that a second variable corresponding to the intersection exit area is true, wherein the preset area is a related area of ​​a lane connected to the intersection exit area;

[0206] The first variable and the second variable corresponding to the intersection exit area are determined as positional relationship information between the object to be predicted and the intersection exit area.

[0207] Solution 7: The method according to Solution 6, after taking each intersection exit area as an intention node in the dynamic Bayesian model, further includes:

[0208] Using the first variable and the second variable corresponding to the exit area of ​​the intersection as the cause argument nodes connecting the intention node;

[0209] The determining the prior probability of the intention node further includes:

[0210] Determining each second conditional probability of the intention node under each cause argument node connected to the intention node;

[0211] The sum of the second conditional probabilities is taken as the first value, the sum of the reference probabilities of the intention nodes at the previous moment is taken as the second value, and the product of the first value and the second value is taken as the prior probability of the intention node at the current moment.

[0212] Solution 8: According to the method of Solution 7, before inputting the position relationship information into a pre-built regional probability prediction model to obtain the predicted heading probability corresponding to the exit area of ​​each intersection, the method further includes:

[0213] Acquire a training data set, wherein the training data set includes sample position relationship information between sample objects and regions where each sample leaves, and a probability label of each region where each sample position relationship information corresponds to;

[0214] Training the regional probability prediction model based on the training data set to obtain a first conditional probability table, a second conditional probability table, and a conversion probability table, or obtaining a first conditional probability model, a second conditional probability model, and a conversion probability model;

[0215] Among them, the first conditional probability table or the first conditional probability model is used to determine each first conditional probability, the second conditional probability table or the second conditional probability model is used to determine each second conditional probability, and the conversion probability table or the conversion probability model is used to determine each conversion probability.

[0216] Solution 9: According to the method of Solution 1, determining the target trajectory corresponding to the object to be predicted based on the target area includes:

[0217] Acquiring state information and environmental information of the object to be predicted at a current moment, wherein the state information includes first position information and first speed information of the object to be predicted, and the environmental information includes second position information and second speed information of other objects to be predicted at the intersection to be passed;

[0218] determining speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment, and determining state information of the object to be predicted at a next moment based on the state information and speed control information of the object to be predicted at the current moment;

[0219] Taking the next moment as the current moment, reacquiring the environmental information at the current moment, and determining the speed control information based on the environmental information and state information at the current moment, until the state information of the object to be predicted at the last moment is determined, and determining the target trajectory corresponding to the object to be predicted based on the state information at each moment;

[0220] The first position information of the object to be predicted at the last moment is the target area.

[0221] Solution 10: The method according to Solution 9, wherein determining the speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment includes:

[0222] Constructing a distance cost function, wherein the distance cost function is used to describe the relationship between the object distance and the cost between the object to be predicted and other objects to be predicted;

[0223] With the goal of minimizing the calculation result of the distance cost function, speed control information corresponding to the minimum distance cost is determined according to the environmental information and state information at the current moment.

[0224] Solution 11: According to the method of Solution 10, constructing the distance cost function includes:

[0225] constructing a relationship between the minimum object distance between the object to be predicted and other objects to be predicted and speed control information according to a change in the square distance between the object to be predicted and other objects to be predicted;

[0226] A distance cost function is constructed according to the relationship between the minimum object distance and the speed control information and a preset first hyperparameter.

[0227] Solution 12: The method according to Solution 11, further comprising: after constructing the distance cost function based on the relationship between the minimum object distance and the speed control information and the preset first hyperparameter;

[0228] If there are multiple other objects to be predicted, then for each of the other objects to be predicted, determine the cost weight corresponding to the other object to be predicted based on the angular difference and distance between the object to be predicted and the other object to be predicted at the current moment, and determine the distance cost component corresponding to the other object to be predicted based on the distance cost function and the cost weight corresponding to the other object to be predicted;

[0229] A final distance cost function is constructed according to the sum of the distance cost components corresponding to the other objects to be predicted.

[0230] Solution 13: The method according to Solution 10, wherein determining the speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment further includes:

[0231] Constructing a speed cost function, wherein the speed cost function is used to describe the relationship between the speed difference between the speed control information of the object to be predicted and the expected speed and the cost;

[0232] A total cost function is constructed based on the speed cost function and the distance cost function, with the goal of minimizing the calculation result of the total cost function. According to the environmental information and state information at the current moment, the speed control information corresponding to the minimum total cost is determined.

[0233] Solution 14: The method according to Solution 13, before constructing the total cost function according to the speed cost function and the distance cost function, further comprising:

[0234] Constructing a directional cost function, wherein the directional cost function is used to describe the relationship between the directional difference between the speed control information of the object to be predicted and the target area and the cost;

[0235] Accordingly, constructing a total cost function according to the speed cost function and the distance cost function includes:

[0236] A total cost function is constructed according to the speed cost function, the direction cost function, and the distance cost function.

[0237] Solution 15: The method according to Solution 9, wherein determining the state information of the object to be predicted at a next moment based on the state information and speed control information of the object to be predicted at a current moment comprises:

[0238] determining the first speed information of the object to be predicted at a next moment according to a preset second hyperparameter, the first speed information of the object to be predicted at a current moment, and the speed control information;

[0239] The first position information of the object to be predicted at the next moment is determined according to the first position information of the object to be predicted at the current moment, the first speed information at the next moment, and the time difference between the current moment and the next moment.

[0240] Solution 16: A device for predicting the behavior of objects at an intersection, comprising:

[0241] An area determination module is used to obtain the exit areas of each intersection corresponding to the intersection to be passed by the current vehicle;

[0242] an information acquisition module, configured to acquire, for each object to be predicted in the intersection to be passed, positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

[0243] A probability prediction module is used to input each of the position relationship information into a pre-built regional probability prediction model to obtain the predicted departure probability corresponding to each of the intersection departure areas;

[0244] A behavior determination module is used to determine a target area corresponding to the object to be predicted in each of the intersection exit areas based on each of the predicted going probabilities, and to determine a target trajectory corresponding to the object to be predicted based on the target area.

[0245] Solution 17. An electronic device, comprising:

[0246] one or more processors;

[0247] a storage device for storing one or more programs;

[0248] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of schemes 1-15.

[0249] Solution 18: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of Solutions 1-15.

[0250] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims

1. A method for predicting the behavior of objects at an intersection, characterized in that: The method comprises: Obtaining each intersection exit area corresponding to the current vehicle's upcoming intersection, wherein the intersection exit area is defined with the starting end of the lane leaving the upcoming intersection as the center, and the intersection exit area is defined with both ends of the sidewalk surrounding the upcoming intersection as the center; For each object to be predicted in the intersection to be passed, obtaining positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located in the intersection to be passed, and the positional relationship information includes a current distance, a current angle difference, and a historical average distance between the object to be predicted and the exit area of ​​the intersection; Inputting each of the position relationship information into a pre-built regional probability prediction model to obtain a predicted going probability corresponding to each intersection exit area, wherein the predicted going probability is the probability that the object to be predicted will leave the intersection to be passed from the intersection exit area; A target area corresponding to the object to be predicted is determined in each of the intersection exit areas based on each of the predicted heading probabilities, and a target trajectory corresponding to the object to be predicted is determined based on the target area.

2. The method according to claim 1, characterized in that The regional probability prediction model is a dynamic Bayesian model, and the inputting of each position relationship information into the pre-built regional probability prediction model to obtain the predicted heading probability corresponding to each intersection exit area includes: For each intersection departure area, the intersection departure area is used as an intention node in the dynamic Bayesian model, and at least one of the current distance, the current angle difference, and the historical average distance corresponding to the intersection departure area is used as each observation argument node connected to the intention node; For each of the intention nodes in the dynamic Bayesian model, the predicted going probability corresponding to the intention node is determined based on the observation argument nodes connected to the intention node.

3. The method according to claim 2, characterized in that The step of determining, for each of the intention nodes in the dynamic Bayesian model, based on the observation evidence nodes connected to the intention node, a predicted going probability corresponding to the intention node includes: For each of the intention nodes, determine the first conditional probability of each of the observation argument nodes under the intention node, the prior probability of the intention node, and the marginal probability of each of the observation argument nodes connected to the intention node; Determine the product of each of the first conditional probabilities and the prior probability, and use the ratio of the product to the edge probability as the predicted going probability corresponding to the intention node.

4. The method according to claim 3, characterized in that Determining the prior probability of the intention node includes: Obtain the predicted going probability corresponding to each intention node at the previous moment, and the conversion probability of each intention node at the previous moment to the intention node at the current moment; For each intention node at the previous moment, the product of the predicted going probability and the conversion probability corresponding to the intention node at the previous moment is taken as the reference probability of the intention node at the previous moment, and the sum of the reference probabilities of each intention node at the previous moment is determined as the prior probability of the intention node at the current moment.

5. The method according to claim 4, characterized in that The obtaining of the positional relationship information between the object to be predicted and each of the exit areas of the intersection further includes: For each of the intersection exit areas, determining whether the intersection exit area is located behind the object to be predicted, and if so, determining that a first variable corresponding to the intersection exit area is a true value; Determining whether the object to be predicted is located within a preset area corresponding to the intersection exit area, and if so, determining that a second variable corresponding to the intersection exit area is true, wherein the preset area is a related area of ​​a lane connected to the intersection exit area; The first variable and the second variable corresponding to the intersection exit area are determined as positional relationship information between the object to be predicted and the intersection exit area.

6. The method according to claim 5, characterized in that After taking each of the intersection exit areas as an intention node in the dynamic Bayesian model, the method further includes: Using the first variable and the second variable corresponding to the exit area of ​​the intersection as the cause argument nodes connecting the intention node; The determining the prior probability of the intention node further includes: Determining each second conditional probability of the intention node under each cause argument node connected to the intention node; The sum of the second conditional probabilities is taken as the first value, the sum of the reference probabilities of the intention nodes at the previous moment is taken as the second value, and the product of the first value and the second value is taken as the prior probability of the intention node at the current moment.

7. The method according to claim 6, characterized in that Before inputting the position relationship information into a pre-built regional probability prediction model to obtain the predicted heading probability corresponding to the exit area of ​​each intersection, the method further includes: Acquire a training data set, wherein the training data set includes sample position relationship information between sample objects and regions where each sample leaves, and a probability label of each region where each sample position relationship information corresponds to; Training the regional probability prediction model based on the training data set to obtain a first conditional probability table, a second conditional probability table, and a conversion probability table, or obtaining a first conditional probability model, a second conditional probability model, and a conversion probability model; Among them, the first conditional probability table or the first conditional probability model is used to determine each first conditional probability, the second conditional probability table or the second conditional probability model is used to determine each second conditional probability, and the conversion probability table or the conversion probability model is used to determine each conversion probability.

8. The method according to claim 1, characterized in that The determining a target trajectory corresponding to the object to be predicted based on the target area includes: Acquiring state information and environmental information of the object to be predicted at a current moment, wherein the state information includes first position information and first speed information of the object to be predicted, and the environmental information includes second position information and second speed information of other objects to be predicted at the intersection to be passed; determining speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment, and determining state information of the object to be predicted at a next moment based on the state information and speed control information of the object to be predicted at the current moment; Taking the next moment as the current moment, reacquiring the environmental information at the current moment, and determining the speed control information based on the environmental information and state information at the current moment, until the state information of the object to be predicted at the last moment is determined, and determining the target trajectory corresponding to the object to be predicted based on the state information at each moment; The first position information of the object to be predicted at the last moment is the target area.

9. The method according to claim 8, characterized in that The determining, based on the environmental information and state information at the current moment, the speed control information of the object to be predicted at the current moment includes: Constructing a distance cost function, wherein the distance cost function is used to describe the relationship between the object distance and the cost between the object to be predicted and other objects to be predicted; With the goal of minimizing the calculation result of the distance cost function, speed control information corresponding to the minimum distance cost is determined according to the environmental information and state information at the current moment.

10. The method according to claim 9, characterized in that The constructing of the distance cost function includes: constructing a relationship between the minimum object distance between the object to be predicted and other objects to be predicted and speed control information according to a change in the square distance between the object to be predicted and other objects to be predicted; A distance cost function is constructed according to the relationship between the minimum object distance and the speed control information and a preset first hyperparameter.

11. The method according to claim 10, characterized in that After constructing the distance cost function according to the relationship between the minimum object distance and the speed control information and the preset first hyperparameter, the method further includes: If there are multiple other objects to be predicted, then for each of the other objects to be predicted, determine the cost weight corresponding to the other object to be predicted based on the angular difference and distance between the object to be predicted and the other object to be predicted at the current moment, and determine the distance cost component corresponding to the other object to be predicted based on the distance cost function and the cost weight corresponding to the other object to be predicted; A final distance cost function is constructed according to the sum of the distance cost components corresponding to the other objects to be predicted.

12. The method according to claim 9, characterized in that The determining of the speed control information of the object to be predicted at the current moment based on the environmental information and state information at the current moment further includes: Constructing a speed cost function, wherein the speed cost function is used to describe the relationship between the speed difference between the speed control information of the object to be predicted and the expected speed and the cost; A total cost function is constructed based on the speed cost function and the distance cost function, with the goal of minimizing the calculation result of the total cost function. According to the environmental information and state information at the current moment, the speed control information corresponding to the minimum total cost is determined.

13. The method according to claim 12, characterized in that Before constructing the total cost function according to the speed cost function and the distance cost function, the method further includes: Constructing a directional cost function, wherein the directional cost function is used to describe the relationship between the directional difference between the speed control information of the object to be predicted and the target area and the cost; Accordingly, constructing a total cost function according to the speed cost function and the distance cost function includes: A total cost function is constructed according to the speed cost function, the direction cost function, and the distance cost function.

14. The method according to claim 8, characterized in that The determining, based on the state information and speed control information of the object to be predicted at the current moment, the state information of the object to be predicted at the next moment includes: determining the first speed information of the object to be predicted at a next moment according to a preset second hyperparameter, the first speed information of the object to be predicted at a current moment, and the speed control information; The first position information of the object to be predicted at the next moment is determined according to the first position information of the object to be predicted at the current moment, the first speed information at the next moment, and the time difference between the current moment and the next moment.

15. A device for predicting the behavior of an object at an intersection, characterized in that: include: an area determination module, configured to obtain each intersection exit area corresponding to the current vehicle's upcoming intersection, wherein the intersection exit area is defined with the starting end of the lane leaving the upcoming intersection as the center, and the intersection exit area is defined with the two ends of the sidewalk surrounding the upcoming intersection as the center; an information acquisition module, configured to acquire, for each object to be predicted at the intersection to be passed, positional relationship information between the object to be predicted and each exit area of ​​the intersection, wherein the object to be predicted is a non-motor vehicle object located at the intersection to be passed, and the positional relationship information includes a current distance, a current angle difference, and a historical average distance between the object to be predicted and the exit area of ​​the intersection; A probability prediction module is used to input each of the position relationship information into a pre-built regional probability prediction model to obtain a predicted going probability corresponding to each intersection exit area, wherein the predicted going probability is the probability that the object to be predicted will leave the intersection to be passed from the intersection exit area; A behavior determination module is used to determine a target area corresponding to the object to be predicted in each of the intersection exit areas based on each of the predicted going probabilities, and to determine a target trajectory corresponding to the object to be predicted based on the target area.

16. An electronic device, characterized in that: The electronic device comprises: one or more processors; a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 14.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 14 is implemented.

Images