Target intention recognition and trajectory prediction method, device, unmanned vehicle and medium

Abstract

The embodiments of the present application relate to the field of unmanned vehicles, and disclose a method, device, unmanned vehicle, and medium for identifying the intention and predicting the trajectory of a target. The method includes: obtaining a number of stop point information and a number of guide lines according to a high-precision map and the current position of the target; selecting the guide lines that meet the target trajectory at the current moment from a number of guide lines according to the historical trajectory of the target as candidate guide lines; using a curve to fit the trajectory based on the historical trajectory and stop point information of the target, and generating a template trajectory corresponding to the maneuvering model; comparing the template trajectory and the candidate guide lines to obtain a distance metric between the template trajectory and the candidate guide lines; obtaining the target intention of the template trajectory according to the distance metric; obtaining a predicted trajectory according to the distance metric and the target intention. The present application can simultaneously and accurately identify the intention and trajectory prediction of the target, and the algorithm is lightweight, the calculation speed is fast, the requirements for high-precision maps are low, and it is easy to implement.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of unmanned driving technology, and in particular to a method, device, unmanned vehicle, and medium for target intention recognition and trajectory prediction. Background Art

[0002] In an unmanned driving system, whether the intention of the targets around the unmanned vehicle can be identified and its future trajectory can be predicted is an important indicator for evaluating the perception ability of the unmanned vehicle and its intelligence. As a successor to the target detection and target tracking functions in the perception module, the target intention recognition and trajectory prediction provide the position information and behavior pattern estimation of the target in the scene at the future moment, providing reliable input to the planning module, making the obstacle avoidance behavior of the unmanned vehicle in complex scenes more intelligent and the road right judgment more accurate.

[0003] In the process of implementing the embodiments of the present application, the inventors of the present application found that in the prior art, the target intention recognition and trajectory prediction are processed separately, the trajectory prediction algorithm mostly relies on the physical motion model of the wooden plaque, the calculation cost is high, the accuracy of long-term prediction is limited, and the separate processing of the target intention recognition and trajectory prediction may lead to the problem of contradictory prediction results. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a method, device, unmanned vehicle and medium for target intention recognition and trajectory prediction, which can simultaneously and accurately recognize the target's intention and predict its trajectory, and has a lightweight algorithm, fast calculation speed, low requirements for high-precision maps, and is easy to implement.

[0005] To solve the above technical problems, the embodiments of the present application adopt the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for target intention recognition and trajectory prediction, including:

[0007] According to the high-precision map and the current position of the target, a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target are obtained in the high-precision map;

[0008] According to the historical trajectory of the target, selecting the guide line that matches the trajectory of the target at the current moment from among the plurality of guide lines as a candidate guide line;

[0009] Based on the historical trajectory of the target and the stopping point information, a curve is used to perform trajectory fitting to generate a template trajectory corresponding to the maneuvering model;

[0010] Comparing the template trajectory with the candidate guide line to obtain a distance metric between the template trajectory and the candidate guide line;

[0011] Obtaining the target intention of the template trajectory according to the distance metric;

[0012] A predicted trajectory is obtained according to the distance metric and the target intention.

[0013] In some embodiments, the step of selecting the guide line that matches the trajectory of the target at the current moment from among the plurality of guide lines as a candidate guide line according to the historical trajectory of the target includes:

[0014] According to the current moment of the target, obtaining the historical state point of the target from the historical trajectory;

[0015] Determine whether the historical state point is on a current guide line of the trajectory of the target at the current moment, the current guide line being at least one of the plurality of guide lines;

[0016] If so, the current guide line is used as a candidate guide line.

[0017] In some embodiments, the step of determining whether the historical state point is on the current guide line of the trajectory of the target at the current moment includes:

[0018] Determine whether the historical state point satisfies preset conditions at the same time, and the preset conditions include: the distance between the historical state point and the current guide line is less than the preset distance; the direction of the current guide line is consistent with the trajectory direction of the historical trajectory in the Frenet coordinate system; the lateral interval of the historical trajectory in the Frenet coordinate system is less than the radial interval.

[0019] In some embodiments, the generating a template trajectory by using a curve to perform trajectory fitting based on the historical trajectory of the target and the stopping point information includes:

[0020] Projecting the historical trajectory onto the Frenet coordinate system where the candidate guide line corresponding to the maneuvering model is located, to obtain an initial point for curve fitting;

[0021] According to the stop point information, obtaining the end point of the curve fitting;

[0022] Based on the initial point and the end point, using a curve to fit a radial trajectory and a lateral trajectory;

[0023] The radial trajectory and the lateral trajectory are combined and saved as a template trajectory.

[0024] In some embodiments, when the maneuvering model is a following maneuvering model or a parking maneuvering model, the generating of a template trajectory by using a curve to perform trajectory fitting based on the historical trajectory of the target and the stopping point information includes:

[0025] When fitting the radial trajectory, if the radial trajectory does not meet the requirements, the radial trajectory that does not meet the requirements is discarded.

[0026] In some embodiments, obtaining the target intention of the template trajectory according to the distance metric includes:

[0027] Classify the intentions of the template trajectories, and obtain the template trajectory with the smallest distance metric among the intentions of each category as the representative template;

[0028] Calculating the probability that the representative template matches the historical trajectory;

[0029] The representative template with the highest probability is taken as the target intention.

[0030] In some embodiments, obtaining a predicted trajectory according to the distance metric and the target intention includes:

[0031] After projecting the historical trajectory into the Frenet coordinate system corresponding to the target intention, a Gaussian process regression model is used to predict the predicted future trajectory of the target intention to obtain a predicted trajectory; wherein the Gaussian process regression model is trained using the historical trajectory as a training sample.

[0032] In a second aspect, an embodiment of the present application further provides a device for target intention recognition and trajectory prediction, the device comprising:

[0033] A guide line acquisition module, used for acquiring a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target in the high-precision map according to the high-precision map and the current position of the target at the target moment;

[0034] A screening module, for screening, according to the historical trajectory of the target, the guide line that matches the trajectory of the target at the current moment from among the plurality of guide lines as a candidate guide line;

[0035] A template trajectory generation module, configured to generate a template trajectory corresponding to the maneuvering model by performing trajectory fitting using a curve based on the historical trajectory of the target and the stopping point information;

[0036] A comparison module, used for comparing the template trajectory with the candidate guide line to obtain a distance measurement between the template trajectory and the candidate guide line;

[0037] A target intention acquisition module, used to obtain the target intention of the template trajectory according to the distance metric;

[0038] The predicted trajectory acquisition module is used to obtain the predicted trajectory according to the distance metric and the target intention.

[0039] In a third aspect, the present application further provides an unmanned vehicle, the unmanned vehicle comprising:

[0040] at least one processor, and

[0041] A memory, wherein the memory is communicatively connected to the processor, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described in the first aspect.

[0042] In a fourth aspect, the present application also provides a non-volatile computer-readable storage medium, characterized in that the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an unmanned vehicle, the unmanned vehicle executes the method described in any one of the first aspects.

[0043] Beneficial effects of the embodiments of the present application: Different from the prior art, the target intention recognition and trajectory prediction method, device, unmanned vehicle and medium provided by the embodiments of the present application, according to the high-precision map and the current position of the target, obtain several stop point information and several guide lines composed of the trajectory points of the target in the high-precision map, and then, according to the historical trajectory of the target, select the guide line that meets the trajectory of the target at the current moment from the several guide line information as the candidate guide line; based on the historical trajectory of the target and the stop point information, use the curve to perform trajectory fitting, and generate a template trajectory corresponding to the maneuvering model. Due to different maneuvering models, the obtained template trajectory is different; compare the template trajectory and the candidate guide line, obtain the distance metric between the template trajectory and the candidate guide line, and obtain the target intention of the template trajectory according to the distance metric, so as to find the maneuvering model that best meets the historical trajectory as the target intention; obtain the predicted trajectory according to the distance metric and the target intention, so as to generate the trajectory prediction in the future time. The present application can simultaneously perform target intention recognition and trajectory prediction, and the algorithm is lightweight, the calculation speed is fast, the requirements for high-precision maps are low, and it is easy to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0045] Figure 1 It is a flowchart of an embodiment of the method for intention recognition and trajectory prediction of the present application;

[0046] Figure 2It is a structural schematic diagram of an embodiment of the intention recognition and trajectory prediction device of the present application;

[0047] Figure 3 It is a schematic diagram of the hardware structure of a controller in one embodiment of the unmanned vehicle of the present application. DETAILED DESCRIPTION

[0048] The present application is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements can also be made without departing from the concept of the present application. These all belong to the protection scope of the present application.

[0049] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0050] It should be noted that, if there is no conflict, the various features in the embodiments of the present application can be combined with each other, all within the scope of protection of the present application. In addition, although the functional module division is performed in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a sequence different from the module division in the device or the flow chart. In addition, the words "first", "second", "third", etc. used herein do not limit the data and execution order, but only distinguish the same items or similar items with basically the same functions and effects.

[0051] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in this specification and in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.

[0052] In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0053] The method and device for target intention recognition and trajectory prediction provided in the embodiment of the present application can be applied to unmanned vehicles. It can be understood that the unmanned vehicle includes a controller and a laser radar. The controller serves as the main control center, and the laser radar is used to obtain point cloud data to obtain the historical trajectory of the target. The controller can simultaneously perform target intention recognition and trajectory prediction, and the algorithm is lightweight, fast in calculation speed, and has low requirements for high-precision maps, making it easy to implement.

[0054] It can be understood that the target involved in this application is other vehicles on the road when the driverless car is driving. Since pedestrians and cyclists are subject to too few rules and constraints in traffic scenarios, have too many intentions, and their future trajectories are too complex to be accurately predicted, and pedestrians and cyclists are the disadvantaged parties in traffic scenarios, more conservative strategies need to be adopted when planning the path of the driverless car. Therefore, the target of this application does not include pedestrians and cyclists.

[0055] See also Figure 1 , is a flow chart of an embodiment of a method for intention recognition and trajectory prediction of a target applied to the present application. The method can be executed by a controller in an unmanned vehicle, and the method includes steps S101 to S106.

[0056] S101: According to a high-precision map and the current position of the target, a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target are obtained in the high-precision map.

[0057] Specifically, the high-precision map can be an offline map, and the high-precision map is used for unmanned vehicle driving, and only the road and its environment are marked, such as lane lines, stop lines, speed limit signs, forks in the road, etc.

[0058] When an unmanned vehicle is driving on a city road, it uses laser radar to obtain surrounding targets, such as a surrounding vehicle, and obtain the current position of the target. Based on the current position of the target, it obtains several guide lines consisting of the target's trajectory points in the high-precision map.

[0059] Furthermore, the plurality of guide lines may be all guide lines within a preset range of the target, and the preset range may be a circular range area with the target as the center and a radius of 200 meters. The obtained guide line information is a set G = {g 1 ,g 2 ,……,g k}, k is the number of guide lines, and each guide line g i It is composed of the line connecting the center lines of the road, and is formed by combining the continuous and directional trajectory points of the target to describe the lane-level route information that the target may follow in the future. t =(x,y,v x ,v y ,a x ,a y ) are the coordinates in the x and y directions, velocity v and acceleration a, s in the Cartesian coordinate system respectively. t is the state of the target at time t. When the current time of the target is T, the earliest state s in the historical trajectory of the target can be T-n+1, exhaustively search for all guide lines within the preset range of the target.

[0060] In addition, several stop point information is obtained from the high-precision map. The stop point is the location point where the vehicle will stop based on the center line of the lane in the traffic scene, usually including intersection stop points and sidewalk stop points. The information of several stop points is represented by a set E, E = {e 1 ,e 2 ,……,e k}, where e i Represents the two-dimensional coordinates of the i-th stop point in the high-precision map.

[0061] S102: According to the historical trajectory of the target, select the guide line that matches the trajectory of the target at the current moment from among the plurality of guide lines as a candidate guide line.

[0062] Specifically, taking the i-th target as an example, the historical trajectory of the i-th target is obtained. The historical trajectory includes the historical state sequence Traj i,T-n:T ={s T-n ,s T-n+1 ,...,s T}, where n represents the length of the state sequence, T represents the timestamp of the current moment, and s t is the state of the target at time t, which is calculated based on the pre-order detection method and tracking algorithm; s t =(x,y,v x ,v y ,a x ,a y ) are the coordinates in the x and y directions, the velocity v and the acceleration a in the Cartesian coordinate system, respectively. n can be set to 10. According to the perception frequency of the unmanned vehicle system of 10 Hz, it can be determined that the historical state sequence contains the motion state of the target in the past 1 second.

[0063] In some embodiments, selecting the guide line that matches the trajectory of the target at the current moment from among the plurality of guide lines as a candidate guide line according to the historical trajectory of the target may include:

[0064] According to the current moment of the target, obtaining the historical state point of the target from the historical trajectory;

[0065] Determining whether the historical state point is on a current guide line, the current guide line being at least one of the plurality of guide lines;

[0066] Determine whether the historical state point is on a current guide line of the trajectory of the target at the current moment, the current guide line being at least one of the plurality of guide lines;

[0067] If so, the current guide line is used as a candidate guide line.

[0068] Specifically, according to the current time T of the target, the earliest state s in the historical trajectory of the target is obtained from the historical trajectory: T-n+1 Then, determine whether the historical state point is on the current guide line of the target's trajectory at the current moment, the current guide line is at least one of the several guide lines, and use the Euclidean distance as a metric to find the distance from each guide line to the historical state point s T-n+1 The nearest state point is taken as the historical state point, and the distance from each guide line to the historical state point is calculated. Then, based on the relationship between the distance from each guide line to the historical state point and different thresholds, it is determined whether the historical state point is on the current guide line of the trajectory of the target at the current moment.

[0069] Further, judging whether the historical state point is on the current guide line of the trajectory of the target at the current moment may include:

[0070] Determine whether the historical state point satisfies preset conditions at the same time, and the preset conditions include: the distance between the historical state point and the current guide line is less than the preset distance; the direction of the current guide line is consistent with the trajectory direction of the historical trajectory in the Frenet coordinate system; the lateral interval of the historical trajectory in the Frenet coordinate system is less than the radial interval.

[0071] The preset distance can be 1.5 meters. If the distance between the historical state point and the current guide line is less than the preset distance of 1.5 meters, and the direction of the current guide line is consistent with the trajectory direction of the historical trajectory in the Frenet coordinate system, and the lateral interval of the historical trajectory in the Frenet coordinate system is less than the radial interval, then it means that the historical state point is on the current guide line and meets the trajectory of the target at the current moment, then the current guide line is used as a candidate guide line.

[0072] It can be understood that the Frenet coordinate system is a coordinate system defined on a trajectory, also known as the sn coordinate system. The s coordinate in the sn coordinate system represents the radial position of the target point on the trajectory, and the n coordinate represents the lateral deviation of the target point equivalent to the trajectory.

[0073] After selecting the guide line that matches the trajectory of the target at the current moment as a candidate guide line, different end points at future moments are selected for different maneuvering models.

[0074] S103: Based on the historical trajectory of the target and the stopping point information, a curve is used to perform trajectory fitting to generate a template trajectory corresponding to the maneuvering model.

[0075] In some embodiments, based on the historical trajectory of the target and the stopping point information, using a curve to perform trajectory fitting to generate a template trajectory may include:

[0076] Projecting the historical trajectory onto the Frenet coordinate system where the candidate guide line corresponding to the maneuvering model is located, to obtain an initial point for curve fitting;

[0077] According to the stop point information, obtaining the end point of the curve fitting;

[0078] Based on the initial point and the end point, using a curve to fit a radial trajectory and a lateral trajectory;

[0079] The radial trajectory and the lateral trajectory are combined and saved as a template trajectory.

[0080] Specifically, first, the historical trajectory is projected onto the Frenet coordinate system where the candidate guide line corresponding to each maneuver model is located. For example, the historical trajectory is represented in Cartesian coordinates as {s T-n ,s T-n+2 ,...,s T The historical trajectory is projected onto the Frenet coordinate system and is expressed as in, is the coordinates of the radial position and lateral position in the Frenet coordinate system, the velocity v and the acceleration a, then, select the initial point of the curve fitting Assuming that the end point of the curve fitting is after T seconds, then the end point of the curve fitting is T seconds, and the state of the end point is Then, the unmanned vehicle samples the state of the end point with a resolution of 1s.

[0081] The maneuvering models include the uniform speed maneuvering model, the vehicle following maneuvering model and the parking maneuvering model.

[0082] For the uniform speed maneuver model, a quartic curve can be used to fit the radial trajectory, and a quintic curve can be used to fit the lateral trajectory.

[0083] The value range of the terminal state time T can be selected as [2,6], and the sampling point can be selected as Among them, s T The equation group is not constrained and can only be calculated by sampling within the time range. Among them, the quartic curve equation of the radial trajectory adopts the following formula:

[0084] s(t)=p s,0 +p s,1 t+p s,2 t 2 +p s,3 t 3 +p s,4 t 4

[0085] Among them, p s ={p s,0 ,p s,1 ,p s,2 ,p s,3 ,p s,4} represents the equation parameters, then the curve equation of radial velocity and acceleration is expressed by the following formula:

[0086] v s (t) = p s,1 +2p s,2 +3p s,3 t 2 +4p s,3 t 4

[0087] a s (t) = 2p s,2 +6p s,3 t+12p s,4 t 2 ;

[0088] Among them, v s (t) represents radial velocity, a s (t) represents acceleration.

[0089] Substitute the curve equations for radial velocity and acceleration into the sample points and sampling points The following system of equations can be obtained:

[0090] s(0)=p s,0

[0091] v s (0) = p s,1

[0092] a s (0) = 2p s,2

[0093] v s (T) = p s,1 +2p s,2 T+3p s,3 T 2 +4p s,4 T 3

[0094] a s (T) = 2p s,2 +6p s,3 T+12p s,4 T 2

[0095] So we can get the equation parameter P s, and then sample the maximum sampling time T at dt=0.1, so as to obtain the temporal sampling of the radial state.

[0096] Correspondingly, the lateral trajectory is fitted using a quintic curve, whose position n(t), velocity v n (t) and acceleration a n (t) can be expressed as:

[0097] n(t)=p n,0 +p n,1 t+p n,2 t 2 +p n,3 t 3 +p n,4 t 4 +p n,5 t 5 ;

[0098] v n (t) = p n,1 +2p n,2 t+3p n,3 t 2 +4p n,4 t 3 +5p n,5 t 4 ;

[0099] a n (t) = 2p n,2 +6p n,3 t+12p n,4 t 2 +20p n,5 t 3 ;

[0100] The position n(t), velocity v n (t) and acceleration a n (t) Substitute the sampling point and sampling points The following system of equations can be obtained:

[0101] y(0)=p n,0

[0102] v y (0) = p n,1

[0103] a y (0) = 2p n,2

[0104] y(T)=p n,0 +p n,1 T+p n,2 T 2 +p n,3T 3 +p n,4 T 4 +p n,5 T 5

[0105] v y (T) = p n,1 +2p n,2 T+3p n,3 T 2 +4p n,4 T 3 +5p n,5 T 4

[0106] a y (T) = 2p n,2 +6p n,3 T+12p n,4 T 2 +20p n,5 T 3

[0107] This results in a lateral trajectory.

[0108] After the radial trajectory and the lateral trajectory are obtained, the radial trajectory and the lateral trajectory are combined and saved as a template trajectory, and the template trajectory corresponds to a uniform speed maneuvering model.

[0109] When the maneuvering model is a following model, similar to the template trajectory of the uniform maneuvering model, the fourth-order curve fitting is used in the radial direction and the fifth-order curve fitting is used in the lateral direction. It is assumed that the front vehicle (the previous target) is d meters in front of the current target in the Frenet coordinate system, and the radial speed is v front,s , the sampling time is t∈[2,6], and the state of the curve fitting termination point is defined as Determine whether the radial trajectory meets the requirements. If the current target speed x t >v front,s T+d, or the current target's speed v t >v x,0 , or the current target velocity v t <v front,s , it is determined that the radial trajectory does not meet the requirements. At this time, the radial trajectory that does not meet the requirements is eliminated, and the remaining trajectory is added to the trajectory template.

[0110] When the maneuver model is a parking maneuver model, the parking maneuver model includes a slow parking model and an emergency braking model. Similar to the template trajectory of the uniform maneuver model, it is assumed that there is a parking point information at a position d meters in front of the current target. The radial direction and lateral direction of the slow parking model are both fitted with a quintic curve. The end point time sampling point T∈[4,10], then the end point state is defined as The emergency braking model is similar to the slow-speed parking model. By assuming that the terminal state acceleration is a large negative acceleration a brake , as the case of emergency braking. The end point time is T∈[1,3], and the sampling end point state is The trajectory filtering method of the parking maneuver model is similar to that of the following maneuver model. The sampled trajectories are filtered using v t >v x,0 , or v t <v front,s Filter out error tracks.

[0111] S104: Compare the template trajectory and the candidate guide line to obtain a distance measure between the template trajectory and the candidate guide line.

[0112] Specifically, after obtaining the template trajectory, the template trajectory is compared with the candidate guide line to obtain a distance metric between the template trajectory and the candidate guide line, which is a Euclidean distance. A template trajectory is Then, the distance metric dist between the template trajectory and the candidate guide line is calculated according to the following formula:

[0113]

[0114] in, and They represent the coordinates of the i-th state in the candidate guide line and the coordinates of the i-th state in the template trajectory respectively.

[0115] S105: Obtaining the target intention of the template trajectory according to the distance metric.

[0116] After obtaining the distance measurement between the template trajectory and the candidate guide line, obtaining the target intention of the template trajectory may include:

[0117] Classify the intentions of the template trajectories, and obtain the template trajectory with the smallest distance metric among the intentions of each category as the representative template;

[0118] Calculating the probability that the representative template matches the historical trajectory;

[0119] The representative template with the highest probability is taken as the target intention.

[0120] Specifically, the intention of the template trajectory can be classified. It can be understood that the intention I includes the route R and the maneuvering model M, and I∈{r,mr∈R,m∈M}, where the route R includes straight going, left turn, right turn, left lane change, and right lane change, which are obtained from the candidate guide lines of the target; the maneuvering model M includes a uniform speed maneuvering model, a following maneuvering model, and a parking maneuvering model, and the parking maneuvering model includes a slow parking model and an emergency braking model.

[0121] The template trajectory with the smallest distance metric in the intent of each category is obtained as the representative template. For example, if the intent category is the candidate guide line for straight travel and the maneuvering model is a uniform speed maneuvering model, then the template trajectory with the smallest distance metric in the template trajectory is selected as the representative template of the intent. Then, the exponential function is used to convert the distance metric into the probability space. Then, the probability that the representative template belongs to the i-th intent is p(I=i):

[0122]

[0123] Therefore, the largest intention in p(I=i) is taken as the target intention.

[0124] S106: Obtain a predicted trajectory according to the distance metric and the target intention.

[0125] In some implementations, obtaining a predicted trajectory according to the distance metric and the target intent may include:

[0126] After projecting the historical trajectory into the Frenet coordinate system corresponding to the target intention, a Gaussian process regression model is used to predict the predicted future trajectory of the target intention to obtain a predicted trajectory; wherein the Gaussian process regression model is trained using the historical trajectory as a training sample.

[0127] Specifically, when using the Gaussian process regression model to optimize the predicted trajectory of the target, first, the historical trajectory is projected into the Frenet coordinate system corresponding to the target intention, and then the Gaussian process regression model is used to predict the predicted future trajectory of the target intention, where the future trajectory includes the relationship between the lateral distance of the trajectory and the time t, to obtain the predicted trajectory.

[0128] Furthermore, in the prior art, Gaussian process regression belongs to semi-supervised learning, which requires using historical trajectories as training samples to train Gaussian process regression parameters, and then deriving a sequence of lateral distances over a period of time in the future, thereby obtaining a more accurate motion trajectory estimate. The Gaussian process regression model can be expressed as:

[0129]

[0130] The dependent variable n(t) represents the lateral distance in the Frenet coordinate system, the independent variable t represents time, and the training sample is the input historical trajectory. In this application, the historical trajectory of the past 1 second is used to predict the trajectory of the next 4 seconds. Therefore, the historical trajectory of 1 second as training data can be expressed as t train =[0,0.1,...,1.0], n train represents the lateral coordinate of the historical trajectory in the Frenet coordinate system. The independent variable of the predicted future trajectory is t pred =[1.1,1.2,...,5.0], predicted dependent variable n pred It represents the result of the lateral coordinate prediction for the next 4 seconds. m(t) represents the mean function, which is defined as a zero-value function, and k(t, t') represents the covariance function, which is defined as a square exponential function:

[0131] m(t)=0

[0132]

[0133] where δ(·) represents the Dirichlet function, σ i , σ n , l represents a hyperparameter.

[0134] Use historical trajectories as training data to calculate the training dependent variable n train The joint probability density mean μ and covariance matrix Σ. For the predicted data t pred , the true value of the prediction result is Then the mean and covariance of the joint probability density are μ * With Σ ** The covariance between the predicted data and the training data dependent variable is Σ * .n train and It conforms to the Gaussian process model, so their combination also conforms to the joint Gaussian distribution

[0135]

[0136] Therefore, the prediction result can be obtained through the marginal probability distribution

[0137]

[0138] Thus obtaining the prediction results.

[0139] In an embodiment of the present application, according to the high-precision map and the current position of the target, several stop point information and several guide lines composed of the trajectory points of the target are obtained in the high-precision map, and then, according to the historical trajectory of the target, the guide lines that meet the trajectory of the target at the current moment are selected from the several guide line information as candidate guide lines; based on the historical trajectory of the target and the stop point information, the curve is used for trajectory fitting to generate a template trajectory corresponding to the maneuvering model. Due to different maneuvering models, different template trajectories are obtained; the template trajectory and the candidate guide line are compared to obtain the distance metric between the template trajectory and the candidate guide line, and the target intent of the template trajectory is obtained according to the distance metric, so as to find the maneuvering model that best meets the historical trajectory as the target intent; according to the distance metric and the target intent, the predicted trajectory is obtained, so as to generate a trajectory prediction in the future time. The present application can simultaneously perform target intent recognition and trajectory prediction, and the algorithm is lightweight, the calculation speed is fast, the requirements for high-precision maps are low, and it is easy to implement.

[0140] The present application also provides a device for target intention recognition and trajectory prediction, see Figure 2 , which shows the structure of a device for identifying the intention and predicting the trajectory of a target provided in an embodiment of the present application. The device 200 for identifying the intention and predicting the trajectory of a target includes:

[0141] A guide line acquisition module 201 is used to acquire a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target in the high-precision map according to the high-precision map and the current position of the target;

[0142] A screening module 202 is used to screen the guide line that matches the track of the target at the current moment from among the plurality of guide lines according to the historical track of the target as a candidate guide line;

[0143] A template trajectory generating module 203, configured to generate a template trajectory corresponding to the maneuvering model by performing trajectory fitting using a curve based on the historical trajectory of the target and the stopping point information;

[0144] A comparison module 204, configured to compare the template trajectory with the candidate guide line to obtain a distance measure between the template trajectory and the candidate guide line;

[0145] A target intention acquisition module 205, used to obtain the target intention of the template trajectory according to the distance metric;

[0146] The predicted trajectory acquisition module 206 is used to obtain a predicted trajectory according to the distance metric and the target intention.

[0147] In an embodiment of the present application, according to the high-precision map and the current position of the target, several stop point information and several guide lines composed of the trajectory points of the target are obtained in the high-precision map, and then, according to the historical trajectory of the target, the guide lines that meet the trajectory of the target at the current moment are selected from the several guide line information as candidate guide lines; based on the historical trajectory of the target and the stop point information, the curve is used for trajectory fitting to generate a template trajectory corresponding to the maneuvering model. Due to different maneuvering models, different template trajectories are obtained; the template trajectory and the candidate guide line are compared to obtain the distance metric between the template trajectory and the candidate guide line, and the target intent of the template trajectory is obtained according to the distance metric, so as to find the maneuvering model that best meets the historical trajectory as the target intent; according to the distance metric and the target intent, the predicted trajectory is obtained, so as to generate a trajectory prediction in the future time. The present application can simultaneously perform target intent recognition and trajectory prediction, and the algorithm is lightweight, the calculation speed is fast, the requirements for high-precision maps are low, and it is easy to implement.

[0148] In some embodiments, the screening module 202 is further configured to:

[0149] According to the current moment of the target, obtaining the historical state point of the target from the historical trajectory;

[0150] Determine whether the historical state point is on a current guide line of the trajectory of the target at the current moment, the current guide line being at least one of the plurality of guide lines;

[0151] If so, the current guide line is used as a candidate guide line.

[0152] In some embodiments, the screening module 202 is further configured to:

[0153] Determine whether the historical state point satisfies preset conditions at the same time, and the preset conditions include: the distance between the historical state point and the current guide line is less than the preset distance; the direction of the current guide line is consistent with the trajectory direction of the historical trajectory in the Frenet coordinate system; the lateral interval of the historical trajectory in the Frenet coordinate system is less than the radial interval.

[0154] In some embodiments, the template trajectory generating module 203 is further used to:

[0155] Projecting the historical trajectory onto the Frenet coordinate system where the candidate guide line corresponding to the maneuvering model is located, to obtain an initial point for curve fitting;

[0156] According to the stop point information, obtaining the end point of the curve fitting;

[0157] Based on the initial point and the end point, using a curve to fit a radial trajectory and a lateral trajectory;

[0158] The radial trajectory and the lateral trajectory are combined and saved as a template trajectory.

[0159] In some embodiments, the target intention recognition and trajectory prediction device 200 includes:

[0160] The rejection module 207 is used to:

[0161] When fitting the radial trajectory, if the radial trajectory does not meet the requirements, the radial trajectory that does not meet the requirements is discarded.

[0162] In some embodiments, the target intention acquisition module 205 is further used to:

[0163] Classify the intentions of the template trajectories, and obtain the template trajectory with the smallest distance metric among the intentions of each category as the representative template;

[0164] Calculating the probability that the representative template matches the historical trajectory;

[0165] The representative template with the highest probability is taken as the target intention.

[0166] In some embodiments, the predicted trajectory acquisition module 206 is further configured to:

[0167] After projecting the historical trajectory into the Frenet coordinate system corresponding to the target intention, a Gaussian process regression model is used to predict the predicted future trajectory of the target intention to obtain a predicted trajectory; wherein the Gaussian process regression model is trained using the historical trajectory as a training sample.

[0168] It should be noted that the above device can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in the device embodiment, please refer to the method provided in the embodiment of the present application.

[0169] Figure 3 FIG. 1 is a schematic diagram of the hardware structure of a controller in an embodiment of an unmanned vehicle. Figure 3 As shown, the controller includes:

[0170] One or more processors 111 and memory 112 . Figure 3 In the figure, a processor 111 and a memory 112 are taken as an example.

[0171] The processor 111 and the memory 112 may be connected via a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0172] The memory 112 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the target intention recognition and trajectory prediction method in the embodiment of the present application (for example, the attached Figure 2 The processor 111 executes various functional applications and data processing of the controller by running the non-volatile software programs, instructions and modules stored in the memory 112, that is, the intention recognition and trajectory prediction method of the target of the above method embodiment is realized.

[0173] The memory 112 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the personnel entry and exit detection device, etc. In addition, the memory 112 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 112 may optionally include a memory remotely arranged relative to the processor 111, and these remote memories may be connected to the unmanned vehicle via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0174] The one or more modules are stored in the memory 112, and when executed by the one or more processors 111, perform the target intention recognition and trajectory prediction method in any of the above method embodiments, for example, perform the above described Figure 1 Steps S101 to S106 of the method; implementing Figure 2 The functions of modules 201-207 in.

[0175] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0176] The embodiment of the present application provides a non-volatile computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors, such as Figure 3 A processor 111 in the embodiment may enable the one or more processors to execute the method for target intention recognition and trajectory prediction in any of the above method embodiments, for example, executing the above described Figure 1Steps S101 to S106 of the method; implementing Figure 2 The functions of modules 201-207 in.

[0177] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0178] Through the description of the above embodiments, a person of ordinary skill in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, or of course by hardware. A person of ordinary skill in the art can understand that all or part of the processes in the above embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may also be combined, the steps may be implemented in any order, and there are many other changes in different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they may still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for target intention recognition and trajectory prediction, It is characterized in that Applied to an unmanned vehicle, the method comprises: According to the high-precision map and the current position of the target, a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target are obtained in the high-precision map; According to the current moment of the target, obtaining the historical state point of the target from the historical trajectory of the target; Determine whether the historical state point is on a current guide line of the trajectory of the target at the current moment, the current guide line being at least one of the plurality of guide lines; If yes, taking the current guide line as a candidate guide line; Based on the historical trajectory of the target and the stopping point information, a curve is used to perform trajectory fitting to generate a template trajectory corresponding to the maneuvering model; Comparing the template trajectory with the candidate guide line to obtain a distance metric between the template trajectory and the candidate guide line; Classify the intentions of the template trajectories, and obtain the template trajectory with the smallest distance metric among the intentions of each category as the representative template; Calculating the probability that the representative template matches the historical trajectory; Taking the representative template with the highest probability as the target intention; Gaussian process regression is used to obtain a predicted trajectory based on the distance metric and the target intent.

2. The method according to claim 1, It is characterized in that The step of determining whether the historical state point is on the current guide line of the trajectory of the target at the current moment includes: Determine whether the historical state point satisfies preset conditions at the same time, and the preset conditions include: the distance between the historical state point and the current guide line is less than the preset distance; the direction of the current guide line is consistent with the trajectory direction of the historical trajectory in the Frenet coordinate system; the lateral interval of the historical trajectory in the Frenet coordinate system is less than the radial interval.

3. The method according to claim 1, It is characterized in that The step of using a curve to perform trajectory fitting based on the historical trajectory of the target and the stopping point information to generate a template trajectory includes: Projecting the historical trajectory onto the Frenet coordinate system where the candidate guide line corresponding to the maneuvering model is located, to obtain an initial point for curve fitting; According to the stop point information, obtaining the end point of the curve fitting; Based on the initial point and the end point, using a curve to fit a radial trajectory and a lateral trajectory; The radial trajectory and the lateral trajectory are combined and saved as a template trajectory.

4. The method according to claim 3, It is characterized in that When the maneuvering model is a following maneuvering model or a parking maneuvering model, after generating a template trajectory by performing trajectory fitting using a curve based on the historical trajectory of the target and the stopping point information, the method further includes: When fitting the radial trajectory, if the radial trajectory does not meet the requirements, the radial trajectory that does not meet the requirements is discarded.

5. The method according to any one of claims 1 to 4, It is characterized in that The step of obtaining a predicted trajectory by using Gaussian process regression according to the distance metric and the target intention includes: After projecting the historical trajectory into the Frenet coordinate system corresponding to the target intention, a Gaussian process regression model is used to predict the predicted future trajectory of the target intention to obtain a predicted trajectory; wherein the Gaussian process regression model is trained using the historical trajectory as a training sample.

6. A device for target intention recognition and trajectory prediction, It is characterized in that The device comprises: A guide line acquisition module, used for acquiring a plurality of stop point information and a plurality of guide lines formed by the trajectory points of the target in the high-precision map according to the high-precision map and the current position of the target at the target moment; A screening module is used to obtain a historical state point of the target from the historical trajectory of the target according to the current moment of the target; determine whether the historical state point is on a current guide line of the trajectory of the target at the current moment, the current guide line being at least one of the plurality of guide lines; if so, use the current guide line as a candidate guide line; A template trajectory generation module, used to generate a template trajectory corresponding to a maneuvering model by using a curve to perform trajectory fitting based on the historical trajectory of the target and the stopping point information; A comparison module, used for comparing the template trajectory with the candidate guide line to obtain a distance measurement between the template trajectory and the candidate guide line; The target intention acquisition module is used to classify the intentions of the template trajectories, obtain the template trajectory with the smallest distance metric among the intentions of each category as the representative template; calculate the probability that the representative template conforms to the historical trajectory; and take the representative template with the largest probability as the target intention; The predicted trajectory acquisition module is used to obtain the predicted trajectory by using Gaussian process regression according to the distance metric and the target intention.

7. An unmanned vehicle, It is characterized in that The unmanned vehicle comprises: at least one processor, and A memory, wherein the memory is communicatively connected to the processor, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of claims 1 to 5.

8. A non-volatile computer-readable storage medium, It is characterized in that The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by the unmanned vehicle, the unmanned vehicle executes the method according to any one of claims 1 to 5.

Images