CN113119996B - Trajectory prediction method and apparatus, electronic device and storage medium - Google Patents
Trajectory prediction method and apparatus, electronic device and storage medium Download PDFInfo
- Publication number
- CN113119996B CN113119996B CN202110298106.4A CN202110298106A CN113119996B CN 113119996 B CN113119996 B CN 113119996B CN 202110298106 A CN202110298106 A CN 202110298106A CN 113119996 B CN113119996 B CN 113119996B
- Authority
- CN
- China
- Prior art keywords
- track
- model
- time period
- moving target
- moving
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
- B60W60/0027—Planning or execution of driving tasks using trajectory prediction for other traffic participants
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The embodiment of the invention is suitable for the technical field of intelligent driving, and provides a track prediction method, a track prediction device, electronic equipment and a storage medium, wherein the track prediction method comprises the following steps: determining a position distribution of the moving object within a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time; predicting the end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions; at least two first trajectories are generated based on the at least two end positions.
Description
Technical Field
The invention relates to the technical field of intelligent driving, in particular to a track prediction method and device, electronic equipment and a storage medium.
Background
The mobile target track prediction is an important component in intelligent driving, and the future track of the mobile target is predicted, so that the intelligent driving system can make a correct decision in advance, and the possibility of accidents in traffic is reduced. At present, in the related art, the driving track of a front vehicle is predicted through a Kalman filtering algorithm according to data received by a current vehicle, but uncertainty of driving motivation and intention of a driver is not considered by the Kalman filtering algorithm, and the accuracy rate of track prediction is low.
Disclosure of Invention
In order to solve the above problem, embodiments of the present invention provide a trajectory prediction method, an apparatus, an electronic device, and a storage medium, so as to at least solve the problem that the accuracy of trajectory prediction of a kalman filter algorithm in the related art is low.
The technical scheme of the invention is realized as follows:
in a first aspect, an embodiment of the present invention provides a trajectory prediction method, where the method includes:
determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time;
predicting the end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions;
at least two first trajectories are generated based on the at least two end positions.
In the foregoing solution, the method further includes:
determining a first probability value for each of the at least two first trajectories; the first probability value represents the similarity degree of a first track and a real moving track of the moving target in the first set time period;
determining a second trajectory from the at least two first trajectories based on the first probability value; the second track represents the predicted moving track of the moving target in a first set time period.
In the foregoing solution, the determining the location distribution of the mobile object within the first set time period based on the first map data includes:
carrying out context coding on the first map data to obtain a coding vector;
inputting the coding vector into a set first model to obtain the position distribution output by the set first model; the set first model is characterized as a multilayer perceptron.
In the foregoing solution, the performing context coding on the first map data to obtain a coding vector includes:
determining at least two fold lines corresponding to the first map data; each of the at least two broken lines represents a vector corresponding to a time point in the second set time period;
and modeling based on the at least two broken lines, wherein the modeled model outputs the coding vector with the context characteristic.
In the foregoing solution, the predicting the end position of the moving target in the first set time period based on the position distribution to obtain at least two end positions includes:
performing feature extraction on the position distribution based on a set second model; the set second model is characterized as a residual error neural network model;
predicting an end position of the moving object in the first set period of time based on the feature extracted from the position distribution.
In the foregoing solution, the generating at least two first tracks based on the at least two end positions includes:
generating at least two cubic spline curves based on the at least two terminal positions and a set cubic spline curve function;
and generating the at least two first tracks corresponding to the at least two cubic spline curves in a region with set grid intervals based on an enumeration method.
In the above solution, the determining a first probability value of each of the at least two first tracks includes:
and inputting all the first tracks in the at least two first tracks into a set classification model to obtain a first probability value of each first track in the at least two first tracks output by the set classification model.
In the above scheme, the method further comprises:
before the set classification model is trained, determining a first parameter of each training data in at least two training data corresponding to the set classification model; the training data represents the real track and the predicted track of the moving target in a third set time period; the first parameter characterizes an average distance between sampling points of the real track and sampling points of the predicted track;
determining a label of corresponding training data based on the first parameter; the labels characterize the accuracy of the predicted trajectories in the corresponding training data.
In the foregoing solution, the method further includes:
updating the set classification model based on an attenuation factor and a second parameter during the training of the set classification model; the second parameter characterizes a ratio of sampling points of the predicted trajectory in the training data outside a movable area of the moving object; the training data represents the real track and the predicted track of the moving target in a third set time period.
In a second aspect, an embodiment of the present invention provides a trajectory prediction apparatus, including:
the position determining module is used for determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represent the moving track of the moving target in a second set time period before the current time;
the terminal prediction module is used for predicting the end point positions of the moving target in the first set time period based on the position distribution to obtain at least two end point positions;
a trajectory generation module for generating at least two first trajectories based on the at least two end positions.
In a third aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the processor and the memory are connected to each other, where the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute the steps of the trajectory prediction method provided in the first aspect of the embodiment of the present invention.
In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, including: the computer-readable storage medium stores a computer program. The computer program, when executed by a processor, performs the steps of the trajectory prediction method as provided by the first aspect of an embodiment of the present invention.
The method and the device for predicting the moving target position of the mobile terminal are characterized by comprising the steps of determining position distribution of the moving target in a first set time period after current time based on first map data, representing a moving track of the moving target in a second set time period before the current time based on the first map data, predicting end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions, and generating at least two first tracks based on the at least two end positions. According to the embodiment of the invention, the position distribution of the moving target is determined through the first map data, the moving track of the moving target is predicted according to the position distribution, the intention of the moving target and the physical constraint of traffic rules are considered, and the accuracy of predicting the moving track is improved.
Drawings
Fig. 1 is a schematic flow chart illustrating an implementation of a trajectory prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating an implementation of another trajectory prediction method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a 2-layer multilayer sensor according to an embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating an implementation flow of another trajectory prediction method according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a ResNet50 network structure according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a cubic spline curve provided by an embodiment of the present invention;
FIG. 7 is a schematic flow chart illustrating an implementation of another trajectory prediction method according to an embodiment of the present invention;
FIG. 8 is a diagram illustrating a trajectory prediction process according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of another trajectory prediction process provided by an embodiment of the present invention;
FIG. 10 is a table illustrating a trajectory prediction capability comparison provided by an embodiment of the present invention;
FIG. 11 is a table illustrating a trajectory prediction capability comparison provided by an embodiment of the present invention;
FIG. 12 is a table illustrating a trajectory prediction capability comparison provided by an embodiment of the present invention;
FIG. 13 is a table illustrating a trajectory prediction capability comparison provided by an embodiment of the present invention;
FIG. 14 is a diagram illustrating experimental results of a trajectory prediction according to an embodiment of the present invention;
FIG. 15 is a schematic diagram of a trajectory prediction device according to an embodiment of the present invention;
fig. 16 is a schematic diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the autonomous driving state, it is important to be able to accurately predict the traveling trajectory of the surrounding vehicles, so that the autonomous driving system can accurately respond to the situation when an emergency occurs. One challenge of autonomous driving is the high uncertainty of future predictions, which contain a large amount of intent information and potential data features.
The related art proposes a method regarding trajectory prediction, and the most classical of them is a method of Kalman filtering (KF, kalman filter). In the autonomous state, the kalman filter may predict the next set of actions to be taken by a vehicle ahead of the autonomous vehicle based on data received by the current vehicle. However, uncertainty of driving motivation and intention of a driver is not considered by the Kalman filter algorithm, the prediction range of the Kalman filter algorithm is too long, and the accuracy is too low. The related art then proposes methods based on data-driven deep learning networks, such as Recurrent Neural Networks (RNNs), variational Automatic Encoders (VAEs), and so on. However, the existing end-to-end deep learning method lacks security guarantee, and the output of the method follows strict traffic constraints, so that the method cannot be integrated into the surrounding physical constraints.
In view of the above disadvantages of the related art, embodiments of the present invention provide a trajectory prediction method, which at least can improve the accuracy of trajectory prediction. In order to explain the technical means of the present invention, the following description will be given by way of specific examples.
Fig. 1 is a schematic flow chart illustrating an implementation process of a trajectory prediction method according to an embodiment of the present invention, where an execution subject of the trajectory prediction method is an electronic device, and the electronic device includes a desktop computer, a notebook computer, a server, and the like. Referring to fig. 1, the trajectory prediction method includes:
s101, determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time.
Here, the second set time period is a history time period before the current time, for example, 5 seconds before the current time. The first set time period is a future time after the current time, for example, the last 5 seconds of the current time.
The moving object may be an object in a moving state such as an automobile or a person. The first map data can represent the moving track of the moving target in the second set time period, the first map data reflects the driving scene of the moving target in the second set time period, the first map data comprises a high-definition map (including lane lines, zebra stripes, traffic lights, speed marks, parking signs and the like) in the environment where the moving target is located, sensor data (such as GPS coordinates and gyroscope data) of the moving target and the like, and the sensor data comprises the driving data of the moving target and the driving data of vehicles around the moving target, such as driving speed, acceleration and the like.
The prediction of the position distribution of the moving object in the first set time period from the first map data may be understood as predicting future coordinates from historical coordinates. Here, the position prediction is performed by using a first model, which is a position prediction model, where the position distribution refers to the position of the moving object at each time point within a first set time period, for example, the position of the moving object within 5s per second. The location may refer to the coordinates of the moving object on a map. For example, a map is converted into an xy plane coordinate system, the position refers to coordinates (x, y) of the moving target on the map, and the position distribution is a set of coordinates of the moving target at each time point in a first set time period.
Referring to fig. 2, in an embodiment, the determining the position distribution of the mobile object within the first set time period based on the first map data includes:
s201, carrying out context coding on the first map data to obtain a coding vector.
The first map data is context-coded, that is, the first map data is vectorized. Here, the first map data is context-coded using a VectorNet model, which is a hierarchical Neural Network (GNN), and is intended to help predict the movement of road users by coding map information by constructing a model. The Vector net model uses a Vector (Vector) to simply express map information and moving objects, and the method breaks away from the traditional picture rendering mode, so that the effects of reducing data volume and calculated amount are achieved.
VectorNet first takes polyline level information and then passes it to the graph to model the high-order interactions between polylines. VectorNet uses a global graph to model each polyline with each other and outputs a coded vector with contextual characteristics.
In an embodiment, said context coding the first map data to obtain a coding vector includes:
the method comprises the following steps: determining at least two fold lines corresponding to the first map data; and each of the at least two broken lines represents a vector corresponding to a time point in the second set time period.
Here, first, the road component members of different local spaces are respectively represented by vectors, the road component members include dynamic traffic participants including vehicles and static road environments including crosswalks, lane boundaries, road signs, and the like.
Representing road components of different local spaces by vectors, e.g., a lane boundary contains a plurality of control points that may constitute a spline curve; the crosswalk may be a polygon defined by several points; the road sign may be represented by one point. Adjacent points are connected to form a vector, and a broken line can be obtained through a vectorization process, wherein the broken line is in one-to-one correspondence with the track and the map elements. All geographic entities may be approximated as polylines defined by a plurality of control points, and dynamic traffic participants may also be approximated as polylines by their motion trajectories, all of which may be represented as a set of vectors.
Here, the point on the same broken line corresponds to a vector at the same time point in the second set time period.
Step two: and modeling based on the at least two broken lines, wherein the modeled model outputs the coding vector with the context characteristic.
Here, the VectorNet model uses a global graph to model each polygonal line with each other, and the modeled model outputs a coded vector having a context characteristic. Contextual characteristics refer to the interaction between multiple traffic participants and their interaction with the road environment.
S202, inputting the coding vector into a set first model to obtain the position distribution output by the set first model; the set first model is characterized as a multilayer perceptron.
Here, the set first model is used to predict the position distribution of the moving object in the first set time period, and the set first model needs to be trained first, and the set first model is trained by using historical map data, which is real data that has already occurred, context-coding the historical map data, and inputting the coding vector and the real position distribution corresponding to the historical map data into the set first model for iterative training.
In practical applications, a predicted moving object is defined
Is the position coordinate (x, y) that may occur within a fixed time frame t.When the position distribution of the moving target in the first set time period is predicted, one position distribution of the moving target can be obtained
The embodiment of the invention carries out position distribution prediction through N discrete sets:
the location distribution is modeled by discrete-continuous factorization:
wherein
Is generalized normal distribution, t refers to time, n refers to the number of coordinates, and x refers to the encoding vector.
Wherein d (-) is implemented by a 2-layer multilayer perceptron (MLP), as shown in fig. 3, fig. 3 is a schematic diagram of a 2-layer multilayer perceptron provided by an embodiment of the present invention, and during the set first model training, the position distribution (x, y) corresponding to the historical map data and the coding vector output by VectorNet are used as the input of the multilayer perceptron d (-) and output as the position distribution and offset of the discrete distribution.
Here, the loss function of the first model is set as:
and training the set first model until the loss function is converged, and finishing the training of the set first model.
And inputting the coding vector into the trained and set first model to obtain the position distribution output by the set first model.
S102, predicting the end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions.
Here, the end position refers to a position of the moving object after the first set time period ends, and the end position is predicted according to the position distribution, so that at least two end positions can be obtained.
Referring to fig. 4, in an embodiment, the predicting the end position of the moving object in the first set time period based on the position distribution to obtain at least two end positions includes:
s401, extracting the characteristics of the position distribution based on a set second model; the set second model is characterized as a residual neural network model.
Here, the set second model is a residual neural network model, and may specifically be a ResNet50 network model, and the ResNet50 network model is improved in the embodiment of the present invention.
Referring to fig. 5, fig. 5 is a schematic structural diagram of a ResNet50 network provided in the embodiment of the present invention, and the embodiment of the present invention makes the following improvements to the ResNet50 network: (1) The filter size of the first convolutional layer is increased to 9*9, so that the purpose of increasing the receptive field is mainly to increase the detection of the travelable area and enhance the learning capability of the network. (2) And reducing the convolution kernel number of the convolution kernels in each residual block to half of that of the original ResNet50 network, wherein the convolution kernel number of the whole network from front to back is 32,64,128,256 in sequence. This is done to reduce the complexity of the model, making the model more lightweight. (3) The 3*3 convolution of the intermediate convolutional layer of the original ResNet50 network is changed to 5*5 convolution, which is done to reduce the number of nonlinear transformations and the complexity of the network. The concatenation of two convolutional layers 3*3 is equivalent to 1 convolutional layer of 5*5, but the convolutional layers of two 3*3 have more nonlinear transformations than the convolutional layers of 1 5*5. Quote two 3*3 convolutional layers use 2 activation functions, while 1 5*5 convolutional layer uses only 1 activation function.
The ResNet50 network model provided by the embodiment of the invention is used for extracting the characteristics of the position distribution, and the extracted characteristics comprise a large amount of characteristic information, such as lane lines, zebra lines, traffic lights, speed marks, parking signs, vehicle sensor data and the like.
S402, predicting the end position of the moving target in the first set time period based on the features extracted from the position distribution.
And predicting the end positions of the moving target in the first set time period according to the features extracted from the position distribution to obtain at least two end positions.
S103, generating at least two first tracks based on the at least two end positions.
The at least two end positions are respectively the track ends of the at least two first tracks, and the at least two first tracks can be generated according to the at least two end positions.
Embodiments of the present invention may use a cubic spline interpolation method to generate the at least two first trajectories. In an embodiment, the generating at least two first trajectories based on the at least two end positions includes:
the method comprises the following steps: and generating at least two cubic spline curves based on the at least two end points and the set cubic spline curve function.
Cubic spline interpolation is a method for generating a motion track of a moving target by using a known path point, and since a track end point and a current position point of the moving target are known, a moving track of the moving target, namely a first track, can be generated according to the cubic spline interpolation method. Interpolation (interpolation) is the solving of curves passing through known points, with known partial data nodes (knots). Spline (spline) is the term for a flexible rule, which is used to connect two adjacent data points in a technical drawing. Spline interpolation is a common interpolation method for obtaining smooth curves in industrial design, and cubic splines are a more extensive one of them.
According to the embodiment of the invention, at least two cubic spline curves are generated by changing the control points of the cubic spline curves according to at least two terminal positions of the moving target and the set cubic spline curve function.
Step two: and generating the at least two first tracks corresponding to the at least two cubic spline curves in a region with set grid intervals based on an enumeration method.
The invention uses a continuous curve to replace the sequence prediction representation of discrete points, so that the representation method is more robust and better reflects a plurality of intentions and movement trends. In the embodiment of the present invention, a cubic spline curve is used to represent the moving track of the moving target, referring to fig. 6, fig. 6 is a schematic diagram of a cubic spline curve provided in the embodiment of the present invention, and the cubic spline curve in fig. 6 has four control points P 0 ，P 1 ，P 2 ，P 3 The set cubic spline curve function is
Where B is the basis function of a cubic spline curve function
According to the historical position coordinates, the current position coordinates and the at least two end positions of the moving target, the embodiment of the invention generates at least two cubic spline curves by changing the control points of the cubic spline curves.
The embodiment of the invention adopts an enumeration method (M region by M) to generate a first track: p M ＝(x+i*ζ，y+j*ζ) i，j∈[-M/2，M/2] Wherein, M is an area with a set grid interval on the map, and ζ is the set grid interval. The embodiment of the invention also sets a curvature parameter theta to generate different curvature points, and the finally generated first track is P f ＝{f(P M θ), where f is a cubic polynomial fitting function, and θ has a range of values { -3, -2, -1,0,1,2,3}.
The method can generate at least two first tracks according to the set cubic spline function and at least two terminal positions. In practical application, the work can be completed by using C + +, the cubic spline interpolation is not required to be realized by the C + +, the related technology has a plurality of mature cubic spline interpolation open source codes, the C + + cubic spline interpolation can be quickly realized by using the codes provided by the related technology, and only the set cubic spline curve function and at least two end points need to be input.
Referring to fig. 7, in an embodiment, the method further comprises:
s701, determining a first probability value of each of the at least two first tracks; the first probability value characterizes how similar a first trajectory is to a real moving trajectory of the moving target within the first set time period.
At least two first tracks are obtained, and each of the at least two first tracks is analyzed to determine the first track which is most likely to be close to the real track of the moving target as only one predicted first track needs to be output finally.
In an embodiment, said determining a first probability value for each of said at least two first tracks comprises:
and inputting all the first tracks in the at least two first tracks into a set classification model to obtain a first probability value of each first track in the at least two first tracks output by the set classification model.
In an embodiment, during the training of the set classification model, the method further includes:
the method comprises the following steps: before the set classification model is trained, determining a first parameter of each training data in at least two training data corresponding to the set classification model; the training data represents the real track and the predicted track of the moving target in a third set time period; the first parameter characterizes an average distance between sampling points of the real trajectory and sampling points of the predicted trajectory.
At least two training data are used for training the set classification model, and each training data comprises a real track and a predicted track of the moving target in a third set time period.
Before training the set classification model, determining a first parameter of each of at least two training data, wherein the first parameter represents an average distance between sampling points of the real track and sampling points of the predicted track. The calculation formula of the first parameter is as follows:
wherein, P t Is a sampling point of the real track, P p Is the sampling point of the predicted trajectory, and M is the number of sampling points. The sampling points refer to coordinate points on a track, and the track is formed by a series of sampling points.
Step two: determining a label of corresponding training data based on the first parameter; the labels characterize the accuracy of the predicted trajectory in the corresponding training data.
Each training data needs to be labeled, and the labels characterize the accuracy of the predicted trajectory in the corresponding training data. For example, when the first parameter is greater than the set value, "bad" is labeled to the corresponding training data, and "bad" indicates that the accuracy of the predicted trajectory is poor.
The embodiment of the invention samples a real track by a point P t And predicted trajectory sampling point P p The average distance between the two serves as a criterion for judging the predicted track:
then, the candidate area is corrected, and the coordinates of the corrected point are used to generate a track: r (x) t -x p ，y-y p ，θ t -θ p )。
In one embodiment, the set loss function of the classification model is:
wherein, C i Is the confidence level of the prediction result for each training trajectory,
are labels of the prediction results of the training trajectories.
And finishing the training of the set classification model when the loss function is converged. Inputting the at least two first tracks into a trained set classification model, and outputting a first probability value of each of the at least two first tracks by the set classification model. Here, the first probability value indicates how close the first trajectory is to the real trajectory of the moving object, and the larger the first probability value is, the more likely the first trajectory is to be close to the real trajectory of the moving object.
In an embodiment, the method further comprises:
updating the set classification model based on an attenuation factor and a second parameter during the training process of the set classification model; the second parameter characterizes a ratio of sampling points of the predicted trajectory in the training data outside a movable area of the moving object; the training data represents the real track and the predicted track of the moving target in a third set time period.
The process of training the model is the process of learning and updating the parameters of the model, and in the embodiment of the invention, the set classification model is updated by using the attenuation factor and the second parameter. The second parameter characterizes a ratio of sample points of the predicted trajectory in the training data outside a movable region of the moving object, the moving object having a corresponding movable region, e.g., if a total of 5 sample points, of which 3 sample points are outside the movable region of the moving object, the second parameter is 3/5.
In the training process of the set classification model, a gradient descent method is usually used to make the set classification model converge, and model parameters are updated continuously in the negative gradient direction, wherein the learning rate is used to determine how large gradient step length is updated each time. In the related art, it is difficult to select a proper learning rate, if the learning rate is too small, the model convergence is slow, and if the learning rate is too large, oscillation around the optimal solution is caused.
In the embodiment of the invention, the learning rate is adjusted by adopting the attenuation factor in the training process of the set classification model. The attenuation factor can be scheduled by the self-adaptive learning rate, the learning rate can be continuously adjusted in the training process, and the learning rate can be reduced in a degressive manner in different stages of training.
The loss function of the classification model set in the embodiment of the present invention is as follows:
wherein the moving target has a corresponding movable region, ε is a ratio of sampling points of the predicted trajectory outside the movable region of the moving target, σ is a decay factor, e is a natural number exponent, g is a classification score,
the classification score added with the attenuation coefficient can obviously improve the accuracy and the efficiency of classification of the classification model.
S702, determining a second trajectory from said at least two first trajectories based on said first probability value; the second track represents the predicted moving track of the moving target in the first set time period.
The first probability value indicates the degree of proximity of the first trajectory to the real trajectory of the moving object, and the larger the first probability value is, the more likely the first trajectory is to be close to the real trajectory of the moving object, and therefore the first trajectory having the largest first probability value is taken as the second trajectory.
The method and the device for predicting the moving target position of the mobile terminal are characterized by comprising the steps of determining position distribution of the moving target in a first set time period after current time based on first map data, representing a moving track of the moving target in a second set time period before the current time based on the first map data, predicting end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions, and generating at least two first tracks based on the at least two end positions. According to the embodiment of the invention, the position distribution of the moving target is determined through the first map data, the moving track of the moving target is predicted according to the position distribution, the intention of the moving target and the physical constraint of traffic rules are considered, and the accuracy of predicting the moving track is improved.
Referring to fig. 8, fig. 8 is a schematic diagram of a trajectory prediction process provided in an application embodiment of the present invention, where the trajectory prediction process is divided into 3 stages, and in the first stage, the map data where the moving object is located is collected, and the map data is context-coded by a VectorNet model to obtain a coding vector, and the position distribution of the moving object is predicted according to the coding vector. And a second stage of predicting the end position of the moving target based on the features extracted from the position distribution by performing feature extraction on the position distribution of the moving target. And the third stage generates at least two moving tracks according to the end position of the moving target, and corrects the at least two moving tracks to obtain a final moving track as a predicted track.
Referring to fig. 9, fig. 9 is a schematic diagram of another trajectory prediction process according to an application embodiment of the present invention, in which map data of a moving target is obtained in a first stage, where the map data includes a high-definition map (including lane lines, zebra stripes, traffic lights, speed markers, parking indicators, and the like) of an environment where the moving target is located, sensor data (such as GPS coordinates and gyroscope data) of the moving target, and the like. And carrying out context coding on the map data through a VectorNet model to obtain a coding vector, and inputting the coding vector into the set first model to obtain the position distribution output by the set first model. And in the second stage, feature extraction is carried out on the position distribution based on a set second model, the set second model is a ResNet50 network model, the end positions of the moving target in the first set time period are predicted according to the features extracted from the position distribution, and at least two end positions are obtained through prediction. And a third stage of generating at least two first tracks according to the at least two end positions, inputting all the first tracks in the at least two first tracks into the set classification model, and obtaining a first probability value of each first track in the at least two first tracks output by the set classification model. The set classification model is a ResNet100 network model, and the first track with the maximum first probability value is used as a second track, and the second track is the predicted track of the moving target.
The application embodiment of the invention comprehensively considers the movement intention of the moving target and the physical constraint of the surrounding traffic rules, and improves the accuracy of the movement track prediction.
The track prediction network provided by the embodiment of the invention is tested on four public data sets of ETH, UCY, apolloCape and Argoverse. ETH and UCY are sets of pedestrian trajectory prediction data, the apollos cape contains the motion trajectory and bird's-eye view coordinates of the target vehicle, and the predicted objects are vehicles, pedestrians, and bicycles/electric vehicles. The argoverte dataset provides a high definition map in addition to the bird's eye view coordinates of the vehicle.
The present invention uses the Average Displacement Error (ADE), the Final Displacement Error (FDE), the Weighted Sum of ADE (WSADE), the Weighted Sum of FDEs (WSFDE), the minimum ADE (minFDE), the minimum FDE (minFDE) and the drivable area Dependency (DAC) to evaluate the proposed algorithm. The present invention compares the proposed trajectory prediction algorithm with the Social LSTM, social GAN, nearest Neighbor (NN) and LSTM ED algorithms.
The effectiveness of the trajectory prediction algorithm in the application embodiment of the present invention was evaluated in fig. 10 and 11 on the ETH, UCY and Apollo data sets with the bird's eye view position coordinates of the target vehicle as input. In order to verify the multi-intention trajectory prediction capability of the algorithm provided by the application embodiment of the invention, an experiment is carried out on the argoverte data set, and the result is shown in fig. 12. In fig. 10, 11 and 12, MITNet (outputs) represents the trajectory prediction algorithm provided by the application embodiment of the present invention, and it can be seen that the trajectory prediction capability of the trajectory prediction algorithm provided by the application embodiment of the present invention is the strongest of several trajectory prediction algorithms, and the trajectory prediction algorithm provided by the application embodiment of the present invention has the smallest error and the smallest dependence on the drivable area.
In addition to the test through the public data set, the test is also carried out through the data set collected by the embodiment of the invention, and the data set of the embodiment of the invention is collected by an automatic driving logistics trolley which is provided with a plurality of RGB cameras and a laser radar and is positioned in the comprehensive technology research and development and test center of the Chinese intelligent vehicle of the maturity in Jiangsu. The present invention uses the 3 second history data to predict the next 5 seconds of vehicle trajectory. The embodiment of the invention extracts about 200000 tracks, and uses 80% of the data to train and 20% of the data to test. The inventive examples also use the same evaluation indices (ADE, FDE, minaDE and DAC) to compare the differences of the proposed method with some reference methods (S-LSTM, S-GAN, LSTM ED). The experimental result of the data set is shown in fig. 13, MITNet (outputs) represents the trajectory prediction algorithm provided by the embodiment of the present invention, and it can be seen that the trajectory prediction capability of the trajectory prediction algorithm provided by the embodiment of the present invention is the minimum error and the minimum dependence of the drivable area in several kinds of trajectory prediction algorithms.
Fig. 14 is a schematic diagram of an experimental result of trajectory prediction provided in an application embodiment of the present invention, and it can be seen from fig. 14 that the trajectory prediction method provided in the embodiment of the present invention can generate an accurate trajectory prediction result.
It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
The technical means described in the embodiments of the present invention may be arbitrarily combined without conflict.
In addition, in the embodiments of the present invention, "first", "second", and the like are used for distinguishing similar objects, and are not necessarily used for describing a specific order or a sequential order.
Referring to fig. 15, fig. 15 is a schematic diagram of a trajectory prediction apparatus according to an embodiment of the present invention, as shown in fig. 15, the apparatus includes: the device comprises a position determining module, a terminal predicting module and a track generating module.
The position determining module is used for determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time;
the terminal prediction module is used for predicting the end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions;
a trajectory generation module for generating at least two first trajectories based on the at least two end positions.
The device further comprises:
a probability determination module for determining a first probability value for each of the at least two first trajectories; the first probability value represents the similarity degree of a first track and a real moving track of the moving target in the first set time period;
a second trajectory determination module for determining a second trajectory from the at least two first trajectories based on the first probability value; the second track represents the predicted moving track of the moving target in a first set time period.
The position determination module is specifically configured to:
carrying out context coding on the first map data to obtain a coding vector;
inputting the coding vector into a set first model to obtain the position distribution output by the set first model; the set first model is characterized as a multilayer perceptron.
The position determination module is specifically configured to:
determining at least two fold lines corresponding to the first map data; each of the at least two broken lines represents a vector corresponding to a time point in the second set time period;
and modeling based on the at least two broken lines, wherein the modeled model outputs the coding vector with the context characteristic.
The terminal prediction module is specifically configured to:
performing feature extraction on the position distribution based on a set second model; the set second model is characterized as a residual error neural network model;
predicting an end position of the moving object in the first set period of time based on the feature extracted from the position distribution.
The trajectory generation module is specifically configured to:
generating at least two cubic spline curves based on the at least two end points and a set cubic spline curve function;
and generating the at least two first tracks corresponding to the at least two cubic spline curves in a region with set grid intervals based on an enumeration method.
The probability determination module is specifically configured to:
and inputting all the first tracks in the at least two first tracks into a set classification model to obtain a first probability value of each first track in the at least two first tracks output by the set classification model.
The device further comprises:
a first parameter determining module, configured to determine, before training the set classification model, a first parameter of each of at least two training data corresponding to the set classification model; the training data represents the real track and the predicted track of the moving target in a third set time period; the first parameter characterizes an average distance between sampling points of the real track and sampling points of the predicted track;
a label determination module for determining a label of the corresponding training data based on the first parameter; the labels characterize accuracy of predicted trajectories in corresponding training data
The device further comprises:
the updating module is used for updating the set classification model based on an attenuation factor and a second parameter in the training process of the set classification model; the second parameter characterizes a ratio of sampling points of the predicted trajectory in the training data outside a movable area of the moving object; the training data represents the real track and the predicted track of the moving target in a third set time period.
In practical applications, the position determining module, the terminal predicting module and the trajectory generating module may be implemented by a Processor in an electronic device, such as a Central Processing Unit (CPU), a Digital Signal Processor (DSP), a Micro Control Unit (MCU), or a Programmable Gate Array (FPGA).
It should be noted that: in the trajectory prediction device provided in the above embodiment, when performing trajectory prediction, only the division of the above modules is used as an example, and in practical applications, the processing may be distributed to different modules according to needs, that is, the internal structure of the device is divided into different modules to complete all or part of the processing described above. In addition, the trajectory prediction device and the trajectory prediction method provided by the above embodiments belong to the same concept, and specific implementation processes thereof are described in the method embodiments and are not described herein again.
Based on the hardware implementation of the program module, in order to implement the method of the embodiment of the present application, an embodiment of the present application further provides an electronic device. Fig. 16 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the present application, and as shown in fig. 16, the electronic device includes:
the communication interface can carry out information interaction with other equipment such as network equipment and the like;
and the processor is connected with the communication interface to realize information interaction with other equipment, and is used for executing the method provided by one or more technical schemes on the electronic equipment side when running a computer program. And the computer program is stored on the memory.
Of course, in practice, the various components in an electronic device are coupled together by a bus system. It will be appreciated that a bus system is used to enable the communication of the connections between these components. The bus system includes a power bus, a control bus, and a status signal bus in addition to a data bus. But for clarity of illustration the various buses are labeled as a bus system in figure 16.
The memory in the embodiments of the present application is used to store various types of data to support the operation of the electronic device. Examples of such data include: any computer program for operating on an electronic device.
It will be appreciated that the memory can be either volatile memory or nonvolatile memory, and can include both volatile and nonvolatile memory. Among them, the nonvolatile Memory may be a Read Only Memory (ROM), a Programmable Read Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a magnetic random access Memory (FRAM), a magnetic random access Memory (Flash Memory), a magnetic surface Memory, an optical Disc, or a Compact Disc Read-Only Memory (CD-ROM); the magnetic surface storage may be disk storage or tape storage. Volatile Memory can be Random Access Memory (RAM), which acts as external cache Memory. By way of illustration and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), synchronous Static Random Access Memory (SSRAM), dynamic Random Access Memory (DRAM), synchronous Dynamic Random Access Memory (SDRAM), double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), enhanced Synchronous Dynamic Random Access Memory (ESDRAM), enhanced Synchronous Dynamic Random Access Memory (Enhanced DRAM), synchronous Dynamic Random Access Memory (SLDRAM), direct Memory (DRmb Access), and Random Access Memory (DRAM). The memories described in the embodiments of the present application are intended to comprise, without being limited to, these and any other suitable types of memory.
The method disclosed in the embodiments of the present application may be applied to a processor, or may be implemented by a processor. The processor may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software. The processor described above may be a general purpose processor, a DSP, or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like. The processor may implement or perform the methods, steps, and logic blocks disclosed in the embodiments of the present application. The general purpose processor may be a microprocessor or any conventional processor or the like. The steps of the method disclosed in the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium located in a memory where a processor reads the programs in the memory and in combination with its hardware performs the steps of the method as previously described.
Optionally, when the processor executes the program, the corresponding process implemented by the electronic device in each method of the embodiment of the present application is implemented, and for brevity, no further description is given here.
In an exemplary embodiment, the present application further provides a storage medium, specifically a computer storage medium, for example, a first memory storing a computer program, where the computer program is executable by a processor of an electronic device to perform the steps of the foregoing method. The computer readable storage medium may be Memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash Memory, magnetic surface Memory, optical disk, or CD-ROM.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus, electronic device and method may be implemented in other ways. The above-described device embodiments are merely illustrative, for example, the division of the unit is only a logical functional division, and there may be other division ways in actual implementation, such as: multiple units or components may be combined, or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the coupling, direct coupling or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between the devices or units may be electrical, mechanical or in other forms.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed on a plurality of network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be separately regarded as one unit, or two or more units may be integrated into one unit; the integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional unit.
Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: a removable storage device, a ROM, a RAM, a magnetic or optical disk, or various other media that can store program code.
Alternatively, the integrated units described above in the present application may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as independent products. Based on such understanding, the technical solutions of the embodiments of the present application or portions thereof that contribute to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a removable storage device, a ROM, a RAM, a magnetic or optical disk, or various other media capable of storing program code.
The technical means described in the embodiments of the present application may be arbitrarily combined without conflict.
In addition, in the examples of the present application, "first", "second", and the like are used for distinguishing similar objects, and are not necessarily used for describing a particular order or sequence.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (11)
1. A trajectory prediction method, characterized in that the method comprises:
determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time;
predicting the end positions of the moving target in the first set time period based on the position distribution to obtain at least two end positions;
generating at least two first trajectories based on the at least two end positions;
wherein the determining of the position distribution of the moving object within a first set time period after the current time based on the first map data comprises:
carrying out context coding on the first map data to obtain a coding vector;
inputting the coding vector into a set first model to obtain the position distribution output by the set first model; the set first model is characterized as a multilayer perceptron.
2. The method of claim 1, further comprising:
determining a first probability value for each of the at least two first tracks; the first probability value represents the similarity degree of a first track and a real moving track of the moving target in the first set time period;
determining a second trajectory from the at least two first trajectories based on the first probability value; the second track represents the predicted moving track of the moving target in a first set time period.
3. The method of claim 1, wherein context coding the first map data to obtain a coded vector comprises:
determining at least two fold lines corresponding to the first map data; each of the at least two broken lines represents a vector corresponding to a time point in the second set time period;
and modeling based on the at least two broken lines, wherein the modeled model outputs the coding vector with the context characteristic.
4. The method of claim 1, wherein predicting the end position of the moving object for the first set period of time based on the position distribution results in at least two end positions, comprising:
performing feature extraction on the position distribution based on a set second model; the set second model is characterized as a residual error neural network model;
predicting an end position of the moving object in the first set period of time based on the feature extracted from the position distribution.
5. The method of claim 1, wherein generating at least two first trajectories based on the at least two end positions comprises:
generating at least two cubic spline curves based on the at least two end points and a set cubic spline curve function;
and generating the at least two first tracks corresponding to the at least two cubic spline curves in a region with set grid intervals based on an enumeration method.
6. The method of claim 2, wherein said determining a first probability value for each of said at least two first tracks comprises:
and inputting all the first tracks in the at least two first tracks into a set classification model to obtain a first probability value of each first track in the at least two first tracks output by the set classification model.
7. The method of claim 6, further comprising:
before the set classification model is trained, determining a first parameter of each training data in at least two training data corresponding to the set classification model; the training data represents the real track and the predicted track of the moving target in a third set time period; the first parameter characterizes an average distance between sampling points of the real track and sampling points of the predicted track;
determining a label of corresponding training data based on the first parameter; the labels characterize the accuracy of the predicted trajectory in the corresponding training data.
8. The method of claim 6, further comprising:
updating the set classification model based on an attenuation factor and a second parameter during the training process of the set classification model; the second parameter characterizes the ratio of sampling points of the predicted track in the training data outside the movable area of the moving target; the training data represents the real track and the predicted track of the moving target in a third set time period.
9. A trajectory prediction device, comprising:
the position determining module is used for determining the position distribution of the moving target in a first set time period after the current time based on the first map data; the first map data represents the moving track of the moving target in a second set time period before the current time;
the terminal prediction module is used for predicting the end point positions of the moving target in the first set time period based on the position distribution to obtain at least two end point positions;
a trajectory generation module for generating at least two first trajectories based on the at least two end positions;
the position determining module is used for carrying out context coding on the first map data to obtain a coding vector; inputting the coding vector into a set first model to obtain the position distribution output by the set first model; the set first model is characterized as a multilayer perceptron.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the trajectory prediction method of any one of claims 1 to 8 when executing the computer program.
11. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions that, when executed by a processor, cause the processor to carry out the trajectory prediction method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110298106.4A CN113119996B (en) | 2021-03-19 | 2021-03-19 | Trajectory prediction method and apparatus, electronic device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110298106.4A CN113119996B (en) | 2021-03-19 | 2021-03-19 | Trajectory prediction method and apparatus, electronic device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113119996A CN113119996A (en) | 2021-07-16 |
| CN113119996B true CN113119996B (en) | 2022-11-08 |
Family
ID=76773621
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110298106.4A Active CN113119996B (en) | 2021-03-19 | 2021-03-19 | Trajectory prediction method and apparatus, electronic device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113119996B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113589338A (en) * | 2021-07-29 | 2021-11-02 | 成都乐动信息技术有限公司 | Method and device for detecting stop point of user in motion process and electronic equipment |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4663726A (en) * | 1985-04-15 | 1987-05-05 | General Electric Co. | Robot control utilizing cubic spline interpolation |
| US10801841B1 (en) * | 2015-10-29 | 2020-10-13 | National Technology & Engineering Solutions Of Sandia, Llc | Trajectory prediction via a feature vector approach |
| EP3437022A4 (en) * | 2017-06-22 | 2019-02-06 | Baidu.com Times Technology (Beijing) Co., Ltd. | Traffic prediction based on map images for autonomous driving |
| CN109583151B (en) * | 2019-02-20 | 2023-07-21 | 阿波罗智能技术(北京)有限公司 | Method and device for predicting running track of vehicle |
| CN111091591B (en) * | 2019-12-23 | 2023-09-26 | 阿波罗智联(北京)科技有限公司 | Collision detection method and device, electronic equipment and storage medium |
| CN111523643B (en) * | 2020-04-10 | 2024-01-05 | 商汤集团有限公司 | Track prediction method, device, equipment and storage medium |
| CN111797780B (en) * | 2020-07-08 | 2022-12-02 | 中国第一汽车股份有限公司 | Vehicle following track planning method, device, server and storage medium |
| CN111931905A (en) * | 2020-07-13 | 2020-11-13 | 江苏大学 | Graph convolution neural network model and vehicle track prediction method using same |
-
2021
- 2021-03-19 CN CN202110298106.4A patent/CN113119996B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113119996A (en) | 2021-07-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113039563B (en) | Learning to generate synthetic data sets for training neural networks | |
| CN112212874B (en) | Vehicle track prediction method and device, electronic equipment and computer readable medium | |
| JP2022516383A (en) | Autonomous vehicle planning | |
| CN111091708A (en) | Vehicle track prediction method and device | |
| CN111782838A (en) | Image question-answering method, image question-answering device, computer equipment and medium | |
| CN112307978B (en) | Target detection method and device, electronic equipment and readable storage medium | |
| JP2022003508A (en) | Trajectory planing model training method and device, electronic apparatus, computer-readable storage medium, and computer program | |
| CN113159403B (en) | Intersection pedestrian track prediction method and device | |
| CN114997307A (en) | Trajectory prediction method, apparatus, device and storage medium | |
| CN115690153A (en) | Intelligent agent track prediction method and system | |
| Yang et al. | Multi-granularity scenarios understanding network for trajectory prediction | |
| Kawasaki et al. | Multimodal trajectory predictions for autonomous driving without a detailed prior map | |
| CN113119996B (en) | Trajectory prediction method and apparatus, electronic device and storage medium | |
| Krüger et al. | Interaction-aware trajectory prediction based on a 3D spatio-temporal tensor representation using convolutional–recurrent neural networks | |
| CN114627441A (en) | Unstructured road recognition network training method, application method and storage medium | |
| Ghanem et al. | An improved and low-complexity neural network model for curved lane detection of autonomous driving system | |
| CN116861262B (en) | Perception model training method and device, electronic equipment and storage medium | |
| Katyal et al. | Occupancy map prediction using generative and fully convolutional networks for vehicle navigation | |
| CN113139696A (en) | Trajectory prediction model construction method and trajectory prediction method and device | |
| CN116071728A (en) | Pedestrian track prediction method based on transducer and attitude estimation and storage medium | |
| Lian et al. | Early intention prediction of pedestrians using contextual attention-based LSTM | |
| CN116080681A (en) | Zhou Chehang identification and track prediction method based on cyclic convolutional neural network | |
| Zhang et al. | Probabilistic trajectory prediction of heterogeneous traffic agents based on layered spatio-temporal graph | |
| Li et al. | Personalized Trajectory Prediction for Driving Behavior Modeling in Ramp-Merging Scenarios | |
| CN116071925B (en) | Track prediction method and device and electronic processing device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |