CN112766561A - Generating type confrontation track prediction method based on attention mechanism - Google Patents
Generating type confrontation track prediction method based on attention mechanism Download PDFInfo
- Publication number
- CN112766561A CN112766561A CN202110053547.8A CN202110053547A CN112766561A CN 112766561 A CN112766561 A CN 112766561A CN 202110053547 A CN202110053547 A CN 202110053547A CN 112766561 A CN112766561 A CN 112766561A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- network
- track
- vector
- pooling
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 230000007246 mechanism Effects 0.000 title claims abstract description 28
- 238000011176 pooling Methods 0.000 claims abstract description 51
- 230000006870 function Effects 0.000 claims abstract description 46
- 230000033001 locomotion Effects 0.000 claims abstract description 16
- 230000003993 interaction Effects 0.000 claims abstract description 15
- 239000013598 vector Substances 0.000 claims description 73
- 238000012549 training Methods 0.000 claims description 66
- 238000009826 distribution Methods 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 19
- 239000010410 layer Substances 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 230000002452 interceptive effect Effects 0.000 claims description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 5
- 230000033228 biological regulation Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000007423 decrease Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 230000003042 antagnostic effect Effects 0.000 claims description 2
- 230000015654 memory Effects 0.000 abstract description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 4
- 230000007787 long-term memory Effects 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- 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
- G06N3/047—Probabilistic or stochastic 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an attention mechanism-based generative confrontation track prediction method, which designs an attention mechanism-based generative confrontation network and trains the generative confrontation network by using an improved loss function. The method comprises the steps of firstly, utilizing an encoder module formed by a long-time memory network LSTM to extract hidden characteristics of pedestrian movement from a pedestrian track, then utilizing a pooling module based on an attention mechanism to distribute influence weights to pedestrians in the same scene so as to fully extract interaction information among the pedestrians, and finally outputting pedestrian track coordinates predicted by a network through a decoder module. The method provided by the invention can improve the prediction precision of the track, can generate a plurality of predicted tracks following the social standard, and can be used in a navigation planning system of the mobile robot, thereby being beneficial to the robot navigation system to plan more reasonable and effective paths in an environment of being merged with people.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to a method for predicting a generating type confrontation track based on an attention mechanism.
Background
The pedestrian trajectory prediction means that the motion trajectory of a pedestrian in a future period of time is predicted according to the motion trajectory of the pedestrian in a past period of time. With the rise of the fields of mobile service robots, automatic driving and the like, the prediction of pedestrian trajectories in dynamic scenes has become a popular research direction. The correct prediction of the pedestrian trajectory is helpful for the intelligent navigation system to plan a more reasonable and effective path. However, the pedestrian trajectory prediction problem is extremely complex, the motion of the pedestrian has certain randomness, and the pedestrian trajectory is subjective and flexible in the decision making process, so that the pedestrian trajectory has the characteristic of diversity. Secondly, in the walking process of the pedestrian, the track of the pedestrian is affected by the surrounding dynamic environment, and the pedestrian usually adjusts the path of the pedestrian according to the common knowledge and social regulations. The above features make the pedestrian trajectory prediction problem challenging.
In the problem of pedestrian trajectory prediction, how to effectively model the interaction between pedestrians is very important for pedestrian trajectory prediction. Most of the existing mainstream methods learn the interaction among pedestrians based on a deep learning technology, so that the pedestrian track is predicted. Among them, the method based on the long and short term memory network LSTM has proven to be very effective in dealing with timing problems, but the method based on LSTM cannot effectively model the spatial relationship between pedestrians. To solve this problem, Alahi et al proposed a Social long-short term memory network (S-LSTM) based on an LSTM network model that predicts a plurality of Social normative trajectories using pooled results by gridding the space and pooling hidden features of pedestrians around each pedestrian according to the grid (see "Human trajectory prediction in crowded spaces, CVPR 2016"). Because the method only models the pedestrian interaction in the local area of the target pedestrian, the interaction of all pedestrians in the scene cannot be simulated efficiently. Gupta et al introduces generative confrontation networks into the pedestrian trajectory prediction problem, proposes a Social confrontation network model, generates a plurality of trajectories meeting Social regulations by performing reverse training on a generator and a discriminator and extracting interactive information of all pedestrians in a scene through a pooling module, and improves prediction accuracy (see 'Social GAN: Social accessible targets with generic adaptive networks, CVPR 2018'). However, in the method, only the spatial position relationship between pedestrians is considered when the mutual information between the pedestrians is extracted, and the influence of factors such as the motion direction and the speed of surrounding pedestrians on the future track of the target pedestrian is ignored, so that the mutual information between the pedestrians cannot be fully extracted. In addition, the method of fighting the network based on the generation formula is easy to cause the phenomenon of imbalance between the strength of the generator and the strength of the discriminator in the network training process, thereby causing the problem of difficult training due to the disappearance of the gradient.
In order to solve the problems, the Guangdong industry university applies for a pedestrian trajectory prediction method based on long-term and short-term memory with the patent number of 202010110743.X, and the invention discloses a pedestrian trajectory prediction method based on long-term and short-term memory, which mainly comprises the following steps: preprocessing the data and converting the data into a matrix of [ the number of pedestrians, 4 ]; an attention mechanism is introduced to select information influencing indexes such as direction, speed and the like when the current pedestrian walks, and all current position information is connected through a full-connection layer; inputting historical state hidden information of global pedestrians in the same scene into a pooling layer for pooling so as to achieve the purpose of sharing the global hidden information; converting the pooling tensor of the historical state hidden information of all the pedestrians in the current state, the position information of the current pedestrian and the information which is selected by the attention mechanism and influences the pedestrians into long-short term memory sequence information through the long-short term memory unit; converting the current state information into a coordinate space through a multi-layer perceptron structure to generate a predicted track sequence:
the patent still has the following defects:
first, in the attention mechanism of the patent, the method for obtaining attention weight only takes into accountThe attention mechanism is obtained by considering the relative position information of the ith pedestrian relative to the jth pedestrian and not comprehensively considering factors such as the speed of the pedestrian j, the moving direction relative to the pedestrian i, the relative distance from the pedestrian i and the like, so that the attention weight is improved by the applicant in the following way: in order to depict the influence of the pedestrian j on the movement of the target pedestrian i, the attention pooling module is used for converting the velocity vector v of the pedestrian jjDistance vector d between pedestrian i and pedestrian jijVelocity vector v of pedestrian iiDistance vector d to pedestrian jijAngle a ofijCosine value cosaijVelocity vector v of pedestrian iiVelocity vector v associated with pedestrian jjAngle b ofijCosine value cosb ofijAre combined into a feature vector qijAnd sending a multilayer full-connection network using a softmax function as an activation function so as to obtain the attention weight of the pedestrian j to the target pedestrian i in the scene.
The method aims at the problem that the training gradient disappears and is difficult to train due to the fact that the strength of a generator and a discriminator in the training process of the traditional GAN network is not matched, noise which is reduced along with time is introduced at the discriminator end in the training process by modifying a loss function, the training effect of the model is improved, and the prediction accuracy of the track is improved. The loss function of the GAN network can be expressed as:
Ltran_GAN=Ex[logD(x)]+Ez[log(1-D(G(z)))]
and the loss function of our improved GAN network is expressed as:
LGAN=Ex[logh(D(x))]+Ez[log(1-h(D(G(z))))]
where h (-) represents a noise function that decreases over time. The advantage of this improvement is that at the initial stage of network training, the intersection of the training data set data distribution and the generator generated data distribution is small, so that the discriminator can easily distinguish between real data and generated data, and the network lacks training gradients. Therefore, in the initial training stage, a certain amount of noise is added to the discriminator so that the training data and the generated data have a certain intersection. With the increase of the training time, the distribution of the data generated by the generator is gradually close to the real data distribution, and the noise is gradually reduced at the moment, so that the network still can be ensured to have a certain training gradient, and the training effect of the network is improved.
Disclosure of Invention
To solve the above existing problems. The invention provides a generating type confrontation track prediction method based on an attention mechanism, which is used for fully extracting interaction information among pedestrians so as to improve the precision of track prediction. If the method is used in a navigation planning system of the service robot, the service robot can plan a more reasonably effective path in a dynamic environment blended with people, and therefore the navigation comfort is improved.
The invention provides an attention mechanism-based generative confrontation track prediction method, which is characterized by comprising the following steps of:
step 1: preprocessing the pedestrian track data and sending the data into an encoder for encoding;
step 2: sending the coded vector to a pooling module based on an attention mechanism for influence weight distribution to obtain a pooling vector;
and step 3: outputting the predicted trajectory of the pedestrian using an LSTM network based decoder;
and 4, step 4: carrying out countermeasure training on the generator and the discriminator by using an Adam algorithm by utilizing the improved loss function;
and 5: and (4) sending the observation track of the pedestrian into a generator of the trained network model to obtain the predicted pedestrian track coordinate.
Further, the encoding processing of the pedestrian trajectory in step 1 includes:
the network receives the historical track of the pedestrian and uses a single-layer full-connection network as an embedded layer to obtain the position change information of the pedestrian i at the time tConversion into a fixed-length feature vectorThen theThe vector is sent into an LSTM network for coding processing, the time sequence characteristics of the track data are learned, and the hidden state of the pedestrian i at the time t is obtained
Where f (-) is an embedded layer using the ReLU activation function, WfAnd WencoderAre the weight parameters of the embedding layer and LSTM network, respectively, and the parameters of the LSTM network are shared by all pedestrians in the scene.
Further, the step 2 of assigning influence weights to pedestrians in the same scene through the attention-based pooling module, and outputting a pooling vector representing interaction information of the pedestrians includes:
in order to depict the influence of a pedestrian j on the motion of a target pedestrian i, a module firstly obtains a pooling vector h in a pooling modeijAnd the velocity vector v of the pedestrian j is calculatedjDistance vector d between pedestrian i and pedestrian jijVelocity vector v of pedestrian iiDistance vector d to pedestrian jijAngle a ofijCosine value cosaijVelocity vector v of pedestrian iiVelocity vector v associated with pedestrian jjAngle b ofijCosine value cosb ofijAre combined into a feature vector qijSending the target pedestrian i into a multilayer fully-connected network using a softmax function as an activation function, and obtaining the attention weight of a pedestrian j to the target pedestrian i in the scene;
then, converging the pooling vectors of all other pedestrians relative to the target pedestrian i in the scene into a final pooling vector HiAnd combining the attention weights of different pedestrians into a weight matrix Watten_iAnd finally, the weight matrix Watten_iAnd pooling of poolsQuantity HiMultiplying to obtain a characteristic vector phiAnd obtaining the pooling vector p of the target pedestrian i in a maximum pooling modeiThe pooling vector represents information needed by the target pedestrian i to make a decision, the information is intuitively understood, and a weight of the influence of all people in a scene on the future track of the target pedestrian i is obtained by adopting an attention mechanism, so that the information p needed by the target pedestrian i to make the decision is summarizediTherefore, the purpose of pedestrian interactive modeling is achieved, and the specific formula is calculated as follows:
qij=[vj,dij,cosaij,cosbij]
qi=[qi1,qi2,...,qij,...,qiN]
Watten_i=s(qi;Ws)
Hi=[hi1,hi2,...,hij,...,hiN]
pi=maxpool(Watten_iHi)
where s (-) denotes a multi-layer fully-connected network using a softmax activation function, WsIs a weight parameter for the network.
Further, the step 3 of outputting the predicted trajectory of the pedestrian by using an LSTM network-based decoder includes:
pooling vectors output by attention pooling moduleHidden vector output by encoder moduleAnd combining the random noise z satisfying the Gaussian distribution into a feature vector as the initial input of the decoderThe decoder firstly converts the position change of the pedestrian at the nearest moment into a feature space through a full-connection network to obtain a feature vectorThen obtaining the current hidden state through the LSTM networkFinally, the predicted track coordinate is obtained by converting the coordinate space through a full-connection networkThe overall calculation formula of the decoder is as follows:
wherein j (-), m (-), and g (-), are all fully connected networks with ReLU activation functions, Wj、WmAnd WgThe weight parameters, W, of the three networks, respectivelydecoderIs the weight parameter of the LSTM network.
Further, the inverse training of the generator and the arbiter using the improved loss function in step 4 comprises:
the network is reversely trained by using an Adam algorithm by utilizing an improved loss function, and the improved loss function mainly comprises two parts, wherein one part is the antagonistic loss L of the GAN networkGANThe other part is the loss of positional offset L between the real trajectory and the predicted trajectory2;
Hypothesis realityThe distribution represented by the training data x of (a) is pdataI.e. x to pdata(x) The generator samples z from the prior noise distribution p, namely z to p (z), and the process of GAN network training is essentially to make the data distribution represented by the output G (z) of the generator as close as possible to the real training set data distribution, and the training loss function L of the traditional GAN networktran_GANExpressed as:
Ltran_GAN=Ex[logD(x)]+Ez[log(1-D(G(z)))]
however, the situation that the generated data of the generator and the real data of the training set can be easily distinguished due to the fact that the distinguishing capability of the discriminator is too strong in the training process of the traditional GAN network, and therefore gradient vanishes and training cannot be conducted is caused, and in order to solve the problem that the training of the traditional GAN network is difficult, step 4 is to apply noise which is reduced along with time to the loss function of the discriminator end in the training process of the GAN network, so that the training data and the generated data have certain intersection, the distribution of the generated data of the generator is gradually close to the distribution of the real data along with the increase of the training time, and at the moment, the noise is gradually reduced, and the network still can have certain training gradient; thus, improved resistance loss LGANExpressed as:
LGAN=Ex[logh(D(x))]+Ez[log(1-h(D(G(z))))]
wherein h (-) represents a noise function that decreases over time;
in order to encourage the network to generate a plurality of tracks meeting the social regulations, the network samples k predicted tracks at each time and selects the track with the minimum position deviation error for calculating the position deviation loss, so the position deviation loss L of the network2Expressed as:
thus, the overall loss function of the network is expressed as:
Ltotal=LGAN+lL2
wherein l is a hyperparameter.
Further, the step 5 of sending the observed trajectory of the pedestrian into the generator to obtain the predicted coordinates of the pedestrian trajectory includes:
and (3) sequentially executing the step (1), the step (2) and the step (3), namely, sending the observation track of the pedestrian into an encoder to carry out encoding processing so as to obtain hidden characteristics of the movement of the pedestrian, extracting interaction information of the pedestrian through an attention pooling module, and finally outputting a predicted track coordinate of the pedestrian through a decoder.
Compared with the prior art, the technical scheme provided by the invention has the following beneficial effects:
1. aiming at the defect that the interactive information among pedestrians cannot be fully extracted by the existing method, an attention pooling module is introduced to associate the moving direction, speed and other elements of the pedestrians with the future tracks of the pedestrians, and influence weight distribution is carried out on the pedestrians in the same scene, so that the interactive information among the pedestrians is more effectively extracted, and the interpretability of the model is improved.
2. Aiming at the problem that the training gradient disappears and is difficult to train due to the fact that the generator and the discriminator are not matched in strength in the training process of the generative countermeasure network, noise which is reduced along with time is introduced at the discriminator end in the training process by modifying the loss function, the training effect of the model is improved, and the prediction precision of the track is improved.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is an overall block diagram of a network model;
FIG. 3 is a schematic view of an attention pooling module;
FIG. 4 is a schematic diagram of a GAN network training process;
fig. 5 is a comparison graph of predicted trajectory visualization.
Detailed Description
The invention is described in further detail below with reference to the following detailed description and accompanying drawings:
the invention provides a generating type confrontation track prediction method based on an attention mechanism, which is used for fully extracting interaction information among pedestrians so as to improve the precision of track prediction. If the method is used in a navigation planning system of the service robot, the service robot can plan a more reasonably effective path in a dynamic environment blended with people, and therefore the navigation comfort is improved.
As shown in fig. 1 and 2, the overall structure of the network model of the present invention mainly includes a generator module and an arbiter module. The generator module is based on an encoder-decoder framework and comprises an encoder, an attention pooling module and a decoder, the generator receives the historical track of the pedestrian, the track of the pedestrian is encoded by the encoder to obtain hidden characteristics of the movement of the pedestrian, then interaction information of the pedestrian is extracted through the pooling module combined with the attention mechanism, and finally the pedestrian position coordinate predicted by the network is output through the decoder module. The discriminator module is mainly composed of an encoder module, receives track input, encodes the track through an encoder, and then scores the truth degree of the track through a classification network.
The method provided by the invention specifically comprises the following steps:
step 1: preprocessing the pedestrian track data and sending the data into an encoder for encoding;
the network receives the historical track of the pedestrian and uses a single-layer full-connection network as an embedded layer to obtain the position change information of the pedestrian i at the time tConversion into a fixed-length feature vectorThen the vector is sent into an LSTM network for coding processing, the time sequence characteristics of the track data are learned, and the hidden state of the pedestrian i at the time t is obtained
Where f (-) is an embedded layer using the ReLU activation function, WfAnd WencoderAre the weight parameters of the embedding layer and LSTM network, respectively, and the parameters of the LSTM network are shared by all pedestrians in the scene.
Step 2: sending the coded vector to a pooling module based on an attention mechanism for influence weight distribution to obtain a pooling vector;
the future trajectory of a pedestrian is always affected by the pedestrian ahead and is related to the speed, direction of movement, relative distance, etc. of these pedestrians. As shown in fig. 3, the future trajectory of the target pedestrian 1 is mainly affected by the pedestrians 2 and 3 in front of the line of sight, which is hardly affected by the pedestrian 4. And the greater the speed of the pedestrian 2, the smaller the relative distance to the pedestrian 1, the greater its effect on the trajectory of the pedestrian 1.
In order to depict the influence of a pedestrian j on the motion of a target pedestrian i, a module firstly obtains a pooling vector h in a pooling modeijAnd the velocity vector v of the pedestrian j is calculatedjDistance vector d between pedestrian i and pedestrian jijVelocity vector v of pedestrian iiDistance vector d to pedestrian jijAngle a ofijCosine value cosaijVelocity vector v of pedestrian iiVelocity vector v associated with pedestrian jjAngle b ofijCosine value cosb ofijAre combined into a feature vector qijAnd sending a multilayer full-connection network using a softmax function as an activation function so as to obtain the attention weight of the pedestrian j to the target pedestrian i in the scene.
Then, all other pedestrians in the scene are pooled relative to the target pedestrian iThe quantities are converged into a final pooled vector HiAnd combining the attention weights of different pedestrians into a weight matrix Watten_i. Finally, the weight matrix Watten_iPooled vector with pooled HiMultiplying to obtain a characteristic vector phiAnd obtaining the pooling vector p of the target pedestrian i in a maximum pooling modei. Intuitively understand that the attention mechanism is adopted to obtain the weight of the future track influence of all people in the scene on the target pedestrian i, so that the information p required by the target pedestrian i to make a decision is summarizediTherefore, the purpose of pedestrian interactive modeling is achieved. The specific formula calculation is as follows:
qij=[vj,dij,cosaij,cosbij]
qi=[qi1,qi2,...,qij,...,qiN]
Watten_i=s(qi;Ws)
Hi=[hi1,hi2,...,hij,...,hiN]
pi=maxpool(Watten_iHi)
where s (-) denotes a multi-layer fully-connected network using a softmax activation function, WsIs a weight parameter for the network.
And step 3: outputting the predicted trajectory of the pedestrian using an LSTM network based decoder;
pooling vectors output by attention pooling moduleHidden vector output by encoder moduleAnd combining the random noise z satisfying the Gaussian distribution into a feature vector as the initial input of the decoderThe decoder first passes through a full connectionThe network converts the position change of the pedestrian at the nearest moment into a feature space to obtain a feature vectorThen obtaining the current hidden state through the LSTM networkFinally, the predicted track coordinate is obtained by converting the coordinate space through a full-connection networkThe overall calculation formula of the decoder is as follows:
wherein j (-), m (-), and g (-), are all fully connected networks with ReLU activation functions, Wj、WmAnd WgThe weight parameters, W, of the three networks, respectivelydecoderIs the weight parameter of the LSTM network.
And 4, step 4: carrying out countermeasure training on the generator and the discriminator by using an Adam algorithm by utilizing the improved loss function;
the improved loss function consists essentially of two parts, one part being the penalty L of the GAN networkGANThe other part is the loss of positional offset L between the real trajectory and the predicted trajectory2。
Suppose that the distribution represented by the real training data x is pdata(i.e., x to p)data(x) The generator samples z (i.e. z-p (z)) from the prior noise distribution p, and the GAN network training process is essentially such that the data distribution represented by the generator output g (z) is as close as possible to the real training set data distribution. Training loss function L of conventional GAN networktran_GANCan be expressed as:
Ltran_GAN=Ex[logD(x)]+Ez[log(1-D(G(z)))]
however, in the training process of the conventional GAN network, the situation that the gradient disappears and training cannot be performed easily is caused because the discrimination capability of the discriminator is too strong and thus the generated data of the generator and the real data of the training set can be easily distinguished.
In order to solve the problem of difficult training of the conventional GAN network, step 4 applies noise which is reduced along with time to the loss function of the GAN network at the discriminator end in the training process, as shown in FIG. 4, wherein a dark solid line represents the data distribution p of the training setdata(x) The light solid line represents the generator generated data distribution pG(z). At the initial stage of network training, the intersection of the two distributions is small, so that the discriminator can easily distinguish real data from generated data, and the network lacks a training gradient. Therefore, in the initial training stage, a certain amount of noise is added to the discriminator so that the training data and the generated data have a certain intersection. With the increase of the training time, the distribution of the data generated by the generator is gradually close to the real data distribution, and the noise is gradually reduced at the moment, so that the network still has a certain training gradient. Thus, the penalty function L proposed hereinGANCan be expressed as:
LGAN=Ex[logh(D(x))]+Ez[log(1-h(D(G(z))))]
where h (-) represents a noise function that decreases over time.
In order to encourage the network to generate a plurality of tracks meeting the social regulations, the network samples k predicted tracks at each time and selects the track with the minimum position deviation error for calculating the position deviation loss, so the position deviation loss L of the network2Can be expressed as:
wherein, YiAndrespectively representing the real trajectory and the predicted trajectory of the pedestrian i.
Thus, the overall loss function of the network can be expressed as:
Ltotal=LGAN+lL2
wherein l is a hyperparameter.
And 5: the observation track of the pedestrian is sent to a generator of the trained network model to obtain a predicted pedestrian track coordinate;
only the step 1, the step 2 and the step 3 need to be executed in sequence, namely, the observation track of the pedestrian is sent into an encoder to be encoded, so that the hidden characteristic of the movement of the pedestrian is obtained, the interaction information of the pedestrian is extracted through an attention pooling module, and finally, the predicted track coordinate of the pedestrian is output through a decoder.
Fig. 5 shows three representative pedestrian trajectory prediction scenarios. In each scene, the left sub-graph represents the real pedestrian motion track, and the right sub-graph represents the observed track and the predicted track of the pedestrian, wherein the solid circles and stars represent the observed track and the predicted track respectively. Therefore, the method provided by the invention can capture the complex interaction between pedestrians accompanied and courtesy mutually, the predicted track of the pedestrian is more consistent with the actual motion scene, and the track predicted by the network is not in conflict with other tracks. Therefore, overall, the predicted track output by the network model provided by the invention not only meets the social specification, but also meets the physical constraint.
TABLE 1 ADE and FDE comparison of different models (t)pred=8/12)
The invention uses the following two indexes to characterize the accuracy of the predicted trajectory.
1) Average offset Error (ADE). And the mean value of Euclidean distances between the predicted track and the real track sequence at each time step is shown.
2) Final Displacement Error (FDE). And representing the Euclidean distance between the predicted track and the real track sequence at the final moment.
The invention selects the most representative Linear, LSTM, S-LSTM and SGAN network models as comparison standards, and the comparison results of various trajectory prediction models are shown in Table 1. Wherein, the data unit in the table is meter, the bold data represents the best result, the attention-GAN is the corresponding network model of the invention, the + DN represents that the attention-GAN introduces the noise which is reduced along with the time in the training process, and the-DN is vice versa.
The data in the comprehensive table can show that the attention pooling mechanism is introduced, so that the information influencing the future track of the target pedestrian can be selectively fused, the model has stronger expressive force, and the interaction of the pedestrian can be accurately depicted. Meanwhile, noise which is reduced along with time is added into the discriminator in the training process, so that the problem that gradient disappears caused by imbalance between the generator and the discriminator can be improved to a certain extent, and the prediction precision of the network is further improved.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, but any modifications or equivalent variations made according to the technical spirit of the present invention are within the scope of the present invention as claimed.
Claims (6)
1. A generating type confrontation track prediction method based on an attention mechanism is characterized by comprising the following steps:
step 1: preprocessing the pedestrian track data and sending the data into an encoder for encoding;
step 2: sending the coded vector to a pooling module based on an attention mechanism for influence weight distribution to obtain a pooling vector;
and step 3: outputting the predicted trajectory of the pedestrian using an LSTM network based decoder;
and 4, step 4: carrying out countermeasure training on the generator and the discriminator by using an Adam algorithm by utilizing the improved loss function;
and 5: and (4) sending the observation track of the pedestrian into a generator of the trained network model to obtain the predicted pedestrian track coordinate.
2. The method for predicting the generative confrontation trajectory based on the attention mechanism as claimed in claim 1, wherein: the encoding processing of the pedestrian track in the step 1 comprises the following steps:
the network receives the historical track of the pedestrian and uses a single-layer full-connection network as an embedded layer to obtain the position change information of the pedestrian i at the time tConversion into a fixed-length feature vectorThen the vector is sent into an LSTM network for coding processing, the time sequence characteristics of the track data are learned, and the hidden state of the pedestrian i at the time t is obtained
Where f (-) is an embedded layer using the ReLU activation function, WfAnd WencoderRespectively an inlay layer and an LSTM meshThe weight parameters of the network, and the parameters of the LSTM network are shared by all pedestrians in the scene.
3. The method for predicting the generative confrontation trajectory based on the attention mechanism as claimed in claim 1, wherein: step 2, the pooling module based on the attention mechanism is used for distributing influence weight to the pedestrians in the same scene, and a pooling vector representing interaction information of the pedestrians is output, and the pooling vector comprises:
in order to depict the influence of a pedestrian j on the motion of a target pedestrian i, a module firstly obtains a pooling vector h in a pooling modeijAnd the velocity vector v of the pedestrian j is calculatedjDistance vector d between pedestrian i and pedestrian jijVelocity vector v of pedestrian iiDistance vector d to pedestrian jijAngle a ofijCosine value cosaijVelocity vector v of pedestrian iiVelocity vector v associated with pedestrian jjAngle b ofijCosine value cosb ofijAre combined into a feature vector qijSending the target pedestrian i into a multilayer fully-connected network using a softmax function as an activation function, and obtaining the attention weight of a pedestrian j to the target pedestrian i in the scene;
then, converging the pooling vectors of all other pedestrians relative to the target pedestrian i in the scene into a final pooling vector HiAnd combining the attention weights of different pedestrians into a weight matrix Watten_iAnd finally, the weight matrix Watten_iPooled vector with pooled HiMultiplying to obtain a characteristic vector phiAnd obtaining the pooling vector p of the target pedestrian i in a maximum pooling modeiThe pooling vector represents information needed by the target pedestrian i to make a decision, the information is intuitively understood, and a weight of the influence of all people in a scene on the future track of the target pedestrian i is obtained by adopting an attention mechanism, so that the information p needed by the target pedestrian i to make the decision is summarizediTherefore, the purpose of pedestrian interactive modeling is achieved, and the specific formula is calculated as follows:
qij=[vj,dij,cosaij,cosbij]
qi=[qi1,qi2,...,qij,...,qiN]
Watten_i=s(qi;Ws)
Hi=[hi1,hi2,...,hij,...,hiN]
pi=maxpool(Watten_iHi)
where s (-) denotes a multi-layer fully-connected network using a softmax activation function, WsIs a weight parameter for the network.
4. The method for predicting the generative confrontation trajectory based on the attention mechanism as claimed in claim 1, wherein: the step 3 of outputting the predicted trajectory of the pedestrian by using an LSTM network-based decoder includes:
pooling vectors output by attention pooling moduleHidden vector output by encoder moduleAnd combining the random noise z satisfying the Gaussian distribution into a feature vector as the initial input of the decoderThe decoder firstly converts the position change of the pedestrian at the nearest moment into a feature space through a full-connection network to obtain a feature vectorThen obtaining the current hidden state through the LSTM networkFinally, the predicted track coordinate is obtained by converting the coordinate space through a full-connection networkThe overall calculation formula of the decoder is as follows:
wherein j (-), m (-), and g (-), are all fully connected networks with ReLU activation functions, Wj、WmAnd WgThe weight parameters, W, of the three networks, respectivelydecoderIs the weight parameter of the LSTM network.
5. The method for predicting the generative confrontation trajectory based on the attention mechanism as claimed in claim 1, wherein: step 4, performing reverse training on the generator and the arbiter by using the improved loss function, comprising:
the network is reversely trained by using an Adam algorithm by utilizing an improved loss function, and the improved loss function mainly comprises two parts, wherein one part is the antagonistic loss L of the GAN networkGANThe other part is the loss of positional offset L between the real trajectory and the predicted trajectory2;
Suppose that the distribution represented by the real training data x is pdataI.e. x to pdata(x) The generator samples z from the prior noise distribution p, i.e. z-p (z), the process of GAN network trainingEssentially, the data distribution represented by the output G (z) of the generator is as close as possible to the real training set data distribution, and the training loss function L of the conventional GAN networktran_GANExpressed as:
Ltran_GAN=Ex[logD(x)]+Ez[log(1-D(G(z)))]
however, the situation that the generated data of the generator and the real data of the training set can be easily distinguished due to the fact that the distinguishing capability of the discriminator is too strong in the training process of the traditional GAN network, and therefore gradient vanishes and training cannot be conducted is caused, and in order to solve the problem that the training of the traditional GAN network is difficult, step 4 is to apply noise which is reduced along with time to the loss function of the discriminator end in the training process of the GAN network, so that the training data and the generated data have certain intersection, the distribution of the generated data of the generator is gradually close to the distribution of the real data along with the increase of the training time, and at the moment, the noise is gradually reduced, and the network still can have certain training gradient; thus, improved resistance loss LGANExpressed as:
LGAN=Ex[logh(D(x))]+Ez[log(1-h(D(G(z))))]
wherein h (-) represents a noise function that decreases over time;
in order to encourage the network to generate a plurality of tracks meeting the social regulations, the network samples k predicted tracks at each time and selects the track with the minimum position deviation error for calculating the position deviation loss, so the position deviation loss L of the network2Expressed as:
thus, the overall loss function of the network is expressed as:
Ltotal=LGAN+lL2
wherein l is a hyperparameter.
6. The method for predicting the generative confrontation trajectory based on the attention mechanism as claimed in claim 1, wherein: and 5, sending the observation track of the pedestrian into a generator to obtain the predicted coordinate of the pedestrian track, wherein the method comprises the following steps:
and (3) sequentially executing the step (1), the step (2) and the step (3), namely, sending the observation track of the pedestrian into an encoder to carry out encoding processing so as to obtain hidden characteristics of the movement of the pedestrian, extracting interaction information of the pedestrian through an attention pooling module, and finally outputting a predicted track coordinate of the pedestrian through a decoder.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110053547.8A CN112766561B (en) | 2021-01-15 | 2021-01-15 | Attention mechanism-based generation type countermeasure track prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110053547.8A CN112766561B (en) | 2021-01-15 | 2021-01-15 | Attention mechanism-based generation type countermeasure track prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112766561A true CN112766561A (en) | 2021-05-07 |
CN112766561B CN112766561B (en) | 2023-11-17 |
Family
ID=75701709
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110053547.8A Active CN112766561B (en) | 2021-01-15 | 2021-01-15 | Attention mechanism-based generation type countermeasure track prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112766561B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113256681A (en) * | 2021-05-26 | 2021-08-13 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on space-time attention mechanism |
CN113269115A (en) * | 2021-06-04 | 2021-08-17 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on Informer |
CN113269114A (en) * | 2021-06-04 | 2021-08-17 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on multiple hidden variable predictors and key points |
CN113538506A (en) * | 2021-07-23 | 2021-10-22 | 陕西师范大学 | Pedestrian trajectory prediction method based on global dynamic scene information depth modeling |
CN113538520A (en) * | 2021-08-02 | 2021-10-22 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and storage medium |
CN113537297A (en) * | 2021-06-22 | 2021-10-22 | 同盾科技有限公司 | Behavior data prediction method and device |
CN113627249A (en) * | 2021-07-05 | 2021-11-09 | 中山大学·深圳 | Navigation system training method and device based on confrontation contrast learning and navigation system |
CN113869170A (en) * | 2021-09-22 | 2021-12-31 | 武汉大学 | Pedestrian trajectory prediction method based on graph division convolutional neural network |
CN113954863A (en) * | 2021-09-07 | 2022-01-21 | 江苏大学 | Vehicle track prediction method based on double-attention machine system and improved Social Gan |
CN113985897A (en) * | 2021-12-15 | 2022-01-28 | 北京工业大学 | Mobile robot path planning method based on pedestrian trajectory prediction and social constraint |
CN114757975A (en) * | 2022-04-29 | 2022-07-15 | 华南理工大学 | Pedestrian trajectory prediction method based on transformer and graph convolution network |
CN116069879A (en) * | 2022-11-14 | 2023-05-05 | 成都信息工程大学 | Method, device, equipment and storage medium for predicting pedestrian track |
CN116663753A (en) * | 2023-08-01 | 2023-08-29 | 江西省供销江南物联网有限公司 | Cold chain food distribution prediction method and system |
CN117273225A (en) * | 2023-09-26 | 2023-12-22 | 西安理工大学 | Pedestrian path prediction method based on space-time characteristics |
CN117332048A (en) * | 2023-11-30 | 2024-01-02 | 运易通科技有限公司 | Logistics information query method, device and system based on machine learning |
CN113985897B (en) * | 2021-12-15 | 2024-05-31 | 北京工业大学 | Mobile robot path planning method based on pedestrian track prediction and social constraint |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781838A (en) * | 2019-10-28 | 2020-02-11 | 大连海事大学 | Multi-modal trajectory prediction method for pedestrian in complex scene |
CN111339867A (en) * | 2020-02-18 | 2020-06-26 | 广东工业大学 | Pedestrian trajectory prediction method based on generation of countermeasure network |
CN111428763A (en) * | 2020-03-17 | 2020-07-17 | 陕西师范大学 | Pedestrian trajectory prediction method based on scene constraint GAN |
CN111461437A (en) * | 2020-04-01 | 2020-07-28 | 北京工业大学 | Data-driven crowd movement simulation method based on generation of confrontation network |
CN111661045A (en) * | 2019-03-05 | 2020-09-15 | 宝马股份公司 | Training a generator unit and a discriminator unit for a trajectory prediction for detecting a collision |
CN111930110A (en) * | 2020-06-01 | 2020-11-13 | 西安理工大学 | Intent track prediction method for generating confrontation network by combining society |
-
2021
- 2021-01-15 CN CN202110053547.8A patent/CN112766561B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111661045A (en) * | 2019-03-05 | 2020-09-15 | 宝马股份公司 | Training a generator unit and a discriminator unit for a trajectory prediction for detecting a collision |
CN110781838A (en) * | 2019-10-28 | 2020-02-11 | 大连海事大学 | Multi-modal trajectory prediction method for pedestrian in complex scene |
CN111339867A (en) * | 2020-02-18 | 2020-06-26 | 广东工业大学 | Pedestrian trajectory prediction method based on generation of countermeasure network |
CN111428763A (en) * | 2020-03-17 | 2020-07-17 | 陕西师范大学 | Pedestrian trajectory prediction method based on scene constraint GAN |
CN111461437A (en) * | 2020-04-01 | 2020-07-28 | 北京工业大学 | Data-driven crowd movement simulation method based on generation of confrontation network |
CN111930110A (en) * | 2020-06-01 | 2020-11-13 | 西安理工大学 | Intent track prediction method for generating confrontation network by combining society |
Non-Patent Citations (2)
Title |
---|
李琳辉等: "基于社会注意力机制的行人轨迹预测方法研究", 通信学报, pages 175 - 183 * |
欧阳俊等: "基于GAN和注意力机制的行人轨迹预测", 激光与光电子学进展, pages 1 - 12 * |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113256681B (en) * | 2021-05-26 | 2022-05-13 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on space-time attention mechanism |
CN113256681A (en) * | 2021-05-26 | 2021-08-13 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on space-time attention mechanism |
CN113269115A (en) * | 2021-06-04 | 2021-08-17 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on Informer |
CN113269114A (en) * | 2021-06-04 | 2021-08-17 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method based on multiple hidden variable predictors and key points |
CN113269115B (en) * | 2021-06-04 | 2024-02-09 | 北京易航远智科技有限公司 | Pedestrian track prediction method based on Informar |
CN113269114B (en) * | 2021-06-04 | 2024-02-02 | 北京易航远智科技有限公司 | Pedestrian track prediction method based on multiple hidden variable predictors and key points |
CN113537297A (en) * | 2021-06-22 | 2021-10-22 | 同盾科技有限公司 | Behavior data prediction method and device |
CN113627249A (en) * | 2021-07-05 | 2021-11-09 | 中山大学·深圳 | Navigation system training method and device based on confrontation contrast learning and navigation system |
CN113538506A (en) * | 2021-07-23 | 2021-10-22 | 陕西师范大学 | Pedestrian trajectory prediction method based on global dynamic scene information depth modeling |
CN114581487B (en) * | 2021-08-02 | 2022-11-25 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method, device, electronic equipment and computer program product |
CN113538520B (en) * | 2021-08-02 | 2022-03-18 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and storage medium |
CN113538520A (en) * | 2021-08-02 | 2021-10-22 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and storage medium |
CN114581487A (en) * | 2021-08-02 | 2022-06-03 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and computer program product |
CN113954863A (en) * | 2021-09-07 | 2022-01-21 | 江苏大学 | Vehicle track prediction method based on double-attention machine system and improved Social Gan |
CN113869170A (en) * | 2021-09-22 | 2021-12-31 | 武汉大学 | Pedestrian trajectory prediction method based on graph division convolutional neural network |
CN113869170B (en) * | 2021-09-22 | 2024-04-23 | 武汉大学 | Pedestrian track prediction method based on graph division convolutional neural network |
CN113985897A (en) * | 2021-12-15 | 2022-01-28 | 北京工业大学 | Mobile robot path planning method based on pedestrian trajectory prediction and social constraint |
CN113985897B (en) * | 2021-12-15 | 2024-05-31 | 北京工业大学 | Mobile robot path planning method based on pedestrian track prediction and social constraint |
CN114757975A (en) * | 2022-04-29 | 2022-07-15 | 华南理工大学 | Pedestrian trajectory prediction method based on transformer and graph convolution network |
CN114757975B (en) * | 2022-04-29 | 2024-04-16 | 华南理工大学 | Pedestrian track prediction method based on transformer and graph convolution network |
CN116069879A (en) * | 2022-11-14 | 2023-05-05 | 成都信息工程大学 | Method, device, equipment and storage medium for predicting pedestrian track |
CN116069879B (en) * | 2022-11-14 | 2023-06-20 | 成都信息工程大学 | Method, device, equipment and storage medium for predicting pedestrian track |
CN116663753A (en) * | 2023-08-01 | 2023-08-29 | 江西省供销江南物联网有限公司 | Cold chain food distribution prediction method and system |
CN116663753B (en) * | 2023-08-01 | 2023-10-20 | 江西省供销江南物联网有限公司 | Cold chain food distribution prediction method and system |
CN117273225B (en) * | 2023-09-26 | 2024-05-03 | 西安理工大学 | Pedestrian path prediction method based on space-time characteristics |
CN117273225A (en) * | 2023-09-26 | 2023-12-22 | 西安理工大学 | Pedestrian path prediction method based on space-time characteristics |
CN117332048A (en) * | 2023-11-30 | 2024-01-02 | 运易通科技有限公司 | Logistics information query method, device and system based on machine learning |
CN117332048B (en) * | 2023-11-30 | 2024-03-22 | 运易通科技有限公司 | Logistics information query method, device and system based on machine learning |
Also Published As
Publication number | Publication date |
---|---|
CN112766561B (en) | 2023-11-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112766561A (en) | Generating type confrontation track prediction method based on attention mechanism | |
Sadeghian et al. | Sophie: An attentive gan for predicting paths compliant to social and physical constraints | |
CN110610271B (en) | Multi-vehicle track prediction method based on long and short memory network | |
US10963738B2 (en) | Method for processing input on basis of neural network learning and apparatus therefor | |
Becker et al. | Red: A simple but effective baseline predictor for the trajnet benchmark | |
Cai et al. | Environment-attention network for vehicle trajectory prediction | |
CN111339867B (en) | Pedestrian trajectory prediction method based on generation of countermeasure network | |
Grigorescu et al. | Neurotrajectory: A neuroevolutionary approach to local state trajectory learning for autonomous vehicles | |
Tagliaferri et al. | Wind direction forecasting with artificial neural networks and support vector machines | |
CN111931902A (en) | Countermeasure network generation model and vehicle track prediction method using the same | |
CN113269363B (en) | Trajectory prediction method, system, equipment and medium of hypersonic aircraft | |
CN115829171B (en) | Pedestrian track prediction method combining space-time information and social interaction characteristics | |
CN112347923A (en) | Roadside end pedestrian track prediction algorithm based on confrontation generation network | |
Wang et al. | A prediction model of vessel trajectory based on generative adversarial network | |
CN112651374B (en) | Future trajectory prediction method based on social information and automatic driving system | |
CN108920805B (en) | Driver behavior modeling system with state feature extraction function | |
Zhang et al. | A bi-directional lstm ship trajectory prediction method based on attention mechanism | |
CN114116944A (en) | Trajectory prediction method and device based on time attention convolution network | |
CN113515131B (en) | Mobile robot obstacle avoidance method and system based on condition variation automatic encoder | |
Akabane et al. | Pedestrian trajectory prediction using pre-trained machine learning model for human-following mobile robot | |
CN117408406A (en) | Planning coupling multi-target vehicle track prediction method and system | |
Zhou et al. | SA-SGAN: A Vehicle Trajectory Prediction Model Based on Generative Adversarial Networks | |
CN114898550B (en) | Pedestrian track prediction method and system | |
CN115700509A (en) | Guideline generation method and system based on simulation feedback data | |
CN113869170B (en) | Pedestrian track prediction method based on graph division convolutional neural network |
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 |