CN111553232B - Gate loop unit network pedestrian trajectory prediction method based on scene state iteration - Google Patents
Gate loop unit network pedestrian trajectory prediction method based on scene state iteration Download PDFInfo
- Publication number
- CN111553232B CN111553232B CN202010319857.5A CN202010319857A CN111553232B CN 111553232 B CN111553232 B CN 111553232B CN 202010319857 A CN202010319857 A CN 202010319857A CN 111553232 B CN111553232 B CN 111553232B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- time
- scene
- gate
- pedestrians
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Psychiatry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Traffic Control Systems (AREA)
Abstract
A gate cycle unit network pedestrian trajectory prediction method based on scene state iteration comprises the steps of data preprocessing, scene state extraction, scene state iteration, prediction model construction and pedestrian trajectory prediction. The method comprises the steps of carrying out coordinate normalization and data enhancement processing on an acquired pedestrian video data set, coding by using a gate cycle unit network, determining the spatial relative position relation between pedestrians in a scene and the attention of the pedestrians in the scene, iterating the acquired hidden layer states of the pedestrians to obtain updated states, constructing a prediction model, repeatedly training the prediction model by using a leave-one-out method, obtaining the optimal training model parameters, applying the optimal training model parameters to the training model, inputting coordinates to be predicted, and predicting the tracks of the pedestrians. The method has the advantages of simplicity, low operation complexity, high prediction accuracy and the like, can be used for predicting the pedestrian movement track by the unmanned vehicle, and can also be used in other technical fields needing to predict the pedestrian movement track.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a method for predicting a walking track of a pedestrian.
Background
With the rapid evolution and development of computer technology, computers are increased in computation speed by geometric multiples. Many sequence-based predictions are also becoming increasingly popular thanks to the dramatic enhancement of computer devices in terms of computational power enhancement. As a typical sequence problem, the pedestrian track has wide application space, such as automatic automobile driving, unmanned delivery robots, real-time traffic monitoring and the like.
The existing pedestrian trajectory prediction method is roughly divided into three types, namely social interaction, motion mode monitoring and deep learning-based method, wherein the social interaction is represented by a social force model, a Gaussian interaction process and the like, the social force model constructs a gravitation-repulsion force model, the Gaussian interaction process introduces a potential interaction network, the two methods simulate the interaction of people in a specific environment, give off an uncommon performance, and have the defect of being limited by a fixed scene. In the aspect of motion mode monitoring, a clustering method is used as a main prediction method, but the method is mainly used for predicting motion mode tracks for avoiding obstacles on static objects, and crowd interaction is not realized. The method mainly applies a cyclic neural network based on a deep learning method, takes the human track as a series of data chains based on inherent time sequence, and is most representative of a Social-LSTM model, which better describes a pedestrian network based on Social state through Social pooling operation, but the state control of pedestrian time steps has limitations and does not show the real-time property of pedestrian state change.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a gate cycle unit network pedestrian trajectory prediction method based on scene state iteration, which is low in operation complexity and high in prediction accuracy.
The technical scheme adopted for solving the technical problems comprises the following steps:
(1) Data pre-processing
Extracting pedestrian trajectory data from the public data set ETH and the data set UCY, including 5 sub-data sets and the nonlinear trajectory of the pedestrian, wherein the two-dimensional coordinates of each pedestrian i at time t are extracted from the data set ETH and the data set UCY and recorded as
All the pedestrian coordinate data are processed by a coordinate normalization and data enhancement method.
(2) Extracting scene states
1) Respectively coding each pedestrian i at a video moment t by using an independent gate cycle unit network, coding all the pedestrians i at the moment t by using the same method, inputting the current position information of the pedestrian i and the hidden layer state of the gate cycle unit network into the gate cycle unit network, and adopting tensor
The position information of the pedestrian i at the time t is determined according to the formula (1):
where phi (-) is a nonlinear function with an offset ReLU, W e Is a weight matrix of the function, b e Is the bias matrix of the function.
2) Determining the hidden layer state of the pedestrian i at the time t according to the formula (2):
wherein the content of the first and second substances,
is the hidden layer state W of the pedestrian i at the time t-1 in the gate cycle unit network G Is an internal weight matrix of the gate-cycle cell network input, b G Is an internal bias matrix for the gate cycle cell network input.
3) Determining the spatial relative position relationship between pedestrians in a scene according to the formula (3)
Wherein phi l Is a ReLU nonlinear function, W l Is the weight of the function and is,
and &>
Representing the spatial coordinates of a pedestrian i and a pedestrian j, respectively, at a time t, i and j being finite positive integers, i ≠ j.
4) Determining feature matrix of relation between pedestrians according to equations (4) and (5)
Wherein sigma F Is a nonlinear function containing a bias Sigmoid, W F Is a weight matrix of the function, b F Is the bias matrix of the function.
5) Determining the attention of a pedestrian in a scene according to equation (6)
Wherein W α Is the weight matrix of the Softmax function.
(3) Iterative scene states
Iterating the states of the hidden layers of the pedestrians obtained by the formula (2) according to the formulas (7) and (8) to obtain updated states
Wherein, denotes a Hadamard product, N is the number of all pedestrians appearing in the scene, is a finite positive integer, W h For the coefficient matrix, z denotes the update gate in the gate cycle unit, σ z Is a Sigmoid nonlinear function in the gate, W z Is a weight matrix of the function, b z Is the functionThe bias matrix of (2).
(4) Building a prediction model
The predicted coordinates of the pedestrian i at the time t +1 are determined according to equation (9)
Where W represents the fully learned parameter matrix.
Combining the coordinate series provided by data set ETH and data set UCY, at time t ob To observe the starting moment, the moment t s For observing the end time, determining a complete observation step length period P:
P＝t s -t ob (10)
all information in the scene state in the step length period P is transmitted to a gate cycle unit network, and the step (3) and the formula (9) of the iterative scene state are adopted to predict the time t s +1 pedestrian coordinate at time t ob +1 as the observation start time, at time t s +1 as the observation end time, the time t being predicted by this method s +2 pedestrian coordinates, continuing to adopt the method to predict until the predicted time t s +t pred The pedestrian coordinates are constructed into a prediction model.
(5) Pedestrian trajectory prediction
Repeatedly training the prediction model by using a leave-one-out method for the data set ETH and the data set UCY, minimizing the mean square error MSE, and determining the mean square error MSE of the predicted track and the real track according to the formula (11):
and obtaining the optimal training model parameters, applying the optimal training model parameters to the training model, inputting coordinate data to be predicted, and predicting the pedestrian track.
In the data preprocessing step (1) of the invention, the coordinate normalization method comprises the following steps: within the observation time length, the initial coordinate of the pedestrian i is taken as the origin. The data enhancement method comprises the following steps: and randomly rotating the video image of the corresponding frame.
In the step (4) of constructing the prediction model, the value range of the observation step length period P is as follows: p is from [5,10 ∈ [ ]](ii) a Said t pred The value range is as follows: t is t pred ∈[8,12]。
The method comprises the steps of carrying out coordinate normalization and data enhancement processing on an acquired pedestrian video data set, coding by using a gate cycle unit network, determining the spatial relative position relation between pedestrians in a scene and the attention of the pedestrians in the scene, iterating the acquired hidden layer states of the pedestrians to obtain updated states, constructing a prediction model, repeatedly training the prediction model by using a leave-one method to obtain the optimal training model parameters, applying the optimal training model parameters to the training model, inputting coordinate data needing to be predicted, and predicting the pedestrian track. The method effectively utilizes the state extracted from the scene and is not limited by the environment, accurately determines the interaction state between the pedestrians in the scene, fully considers the interaction between the pedestrians in the scene, and corrects the hidden layer state of the pedestrians in real time according to the interaction result between the pedestrians to obtain the predicted track of the pedestrians, wherein the predicted track is close to the real track of the pedestrians. The method has the advantages of simplicity, low operation complexity, high prediction accuracy and the like, can be used for predicting the motion trail of the pedestrian by the unmanned vehicle, can avoid the pedestrian by the unmanned vehicle, and can also be used in other technical fields needing to predict the motion trail of the pedestrian.
Drawings
FIG. 1 is a flowchart of example 1 of the present invention.
Detailed Description
The present invention will be described in further detail below with reference to the drawings and examples, but the present invention is not limited to the embodiments described below.
Example 1
Taking 3 video data sets from the disclosed video data set ETH, taking 2 video data sets from the video data set UCY as an example, the gate cycle unit network pedestrian trajectory prediction method based on scene state iteration of the present embodiment has the following steps (see fig. 1):
(1) Data pre-processing
Taking 3 video data sets from the public video data set ETH, taking 2 video data sets from the video data set UCY, extracting pedestrian trajectory data, including 5 sub data sets and nonlinear trajectory of pedestrian, wherein two-dimensional coordinates of each pedestrian i at time t are extracted from the data set ETH and the data set UCY and recorded as
All the pedestrian coordinate data are processed by a coordinate normalization and data enhancement method.
The coordinate normalization method of the embodiment comprises the following steps: within the observation time length, the initial coordinate of the pedestrian i is taken as the origin. The data enhancement method comprises the following steps: and randomly rotating the video image of the corresponding frame.
(2) Extracting scene states
1) Respectively coding each pedestrian i at a video moment t by using an independent gate cycle unit network, coding all the pedestrians i at the moment t by using the same method, inputting the current position information of the pedestrian i and the hidden layer state of the gate cycle unit network into the gate cycle unit network, and adopting tensor
The position information of the pedestrian i at the time t is determined according to the formula (1):
where φ is a nonlinear function with offset ReLU, W e Is a weight matrix of the function, b e Is the bias matrix of the function.
2) Determining the hidden layer state of the pedestrian i at the time t according to the formula (2):
wherein the content of the first and second substances,
is the hidden layer state W of the pedestrian i at the time t-1 in the gate cycle unit network G Is an internal weight matrix of the gate-cycle cell network input, b G Is an internal bias matrix for the gate cycle cell network input.
3) Determining the spatial relative position relationship between pedestrians in a scene according to the formula (3)
Wherein phi l Is a ReLU nonlinear function, W l Is the weight of the function and is,
and &>
Respectively representing the spatial coordinates of a pedestrian i and a pedestrian j at a time t, i and j being finite positive integers, i ≠ j.
4) Determining feature matrix of relation between pedestrians according to equations (4) and (5)
Wherein σ F Is containing an offset Sigmoid nonlinear function, W F Is a weight matrix of the function, b F Is the bias matrix of the function.
5) Determining the attention of a pedestrian in a scene according to equation (6)
Wherein W α Is the weight matrix of the Softmax function.
The step fully considers the interaction between pedestrians in the scene, obtains the pedestrian relationship characteristics and the attention of the pedestrians by the current extraction of the hidden states of the pedestrians and the utilization of the space relative position relationship between the pedestrians, and further describes the behavior characteristics of the pedestrians in the scene.
(3) Iterative scene states
Iterating the pedestrian hidden layer state obtained by the formula (2) according to the formulas (7) and (8) to obtain an updated hidden layer state
Wherein, denotes a Hadamard product, N is the number of all pedestrians appearing in the scene, is a finite positive integer, W h For the coefficient matrix, z denotes the update gate in the gate cycle unit, σ z Is a Sigmoid nonlinear function in the gate, W z Is a weight matrix of the function, b z Is the bias matrix of the function.
In the step, through a method of iterating a scene state, the interaction result between pedestrians is displayed in a mode of correcting the hidden layer state of the pedestrian i at the current time t, and the real-time property of reflecting the pedestrian state change in the scene through the method is reflected.
(4) Building a prediction model
The predicted coordinates of the pedestrian i at the time t +1 are determined according to equation (9)
Where W represents the fully learned parameter matrix.
Combining the coordinate series provided by data set ETH and data set UCY, at time t ob To observe the starting moment, the moment t s For observing the end time, determining a complete observation step length period P:
P＝t s -t ob (10)
all information in the scene state in the step length period P is transmitted to the gate cycle unit network, the value P of the observation step length period P in the embodiment is 8, and the time t is predicted by adopting the iterative scene state step (3) and the formula (9) s +1 pedestrian coordinate at time t ob +1 as the observation start time, at time t s +1 as the observation end time, and the time t is predicted by this method s +2 pedestrian coordinates, continuing to adopt the method to predict until the predicted time t s +t pred Pedestrian coordinates of (1), t of the present embodiment pred The value range of (2) is 10, and a prediction model is constructed.
(5) Pedestrian trajectory prediction
Repeatedly training the prediction model by using a leave-one-out method for the data set ETH and the data set UCY, minimizing the mean square error MSE, and determining the mean square error MSE of the predicted track and the real track according to the formula (11):
and obtaining the optimal training model parameters, applying the optimal training model parameters to the training model, inputting coordinate data to be predicted, and predicting the pedestrian track.
Example 2
Taking 3 video data sets from the disclosed video data set ETH, taking 2 video data sets from the video data set UCY as an example, the gate cycle unit network pedestrian trajectory prediction method based on scene state iteration of the present embodiment has the following steps:
(1) Data pre-processing
This procedure is the same as in example 1.
(2) Extracting scene states
This procedure is the same as in example 1.
(3) Iterative scene states
This procedure is the same as in example 1.
(4) Building a prediction model
The predicted coordinates of the pedestrian i at the time t +1 are determined according to the equation (9)
Where W represents the fully learned parameter matrix.
Combining the coordinate series provided by data set ETH and data set UCY, at time t ob To observe the starting moment, the moment t s For observing the end time, determining a complete observation step length period P:
P＝t s -t ob (10)
all information in the scene state in the step length period P is transmitted to the gate cycle unit network, the value P of the observation step length period P in the embodiment is 5, and the time t is predicted by adopting the iterative scene state step (3) and the formula (9) s +1 pedestrian coordinate at time t ob +1 as the observation start time, at time t s +1 makingTo observe the end time, the time t is predicted by the method s +2 pedestrian coordinates, continuing to adopt the method to predict until the predicted time t s +t pred Pedestrian coordinates of, t of the embodiment pred The value of (2) is 8, and a prediction model is constructed.
The other steps were the same as in example 1.
Example 3
Taking 3 video data sets from the disclosed video data set ETH, taking 2 video data sets from the video data set UCY as an example, the gate cycle unit network pedestrian trajectory prediction method based on scene state iteration of the present embodiment has the following steps:
(1) Data pre-processing
This procedure is the same as in example 1.
(2) Extracting scene states
This procedure is the same as in example 1.
(3) Iterative scene states
This procedure is the same as in example 1.
(4) Building a prediction model
The predicted coordinates of the pedestrian i at the time t +1 are determined according to equation (9)
Where W represents the fully learned parameter matrix.
Combining the coordinate series provided by data set ETH and data set UCY, at time t ob To observe the starting moment, the moment t s For observing the end time, determining a complete observation step length period P:
P＝t s -t ob (10)
all information in the scene state in the step length period P is transmitted to a gate cycle unit network, the value P of the observation step length period P in the embodiment is 10, and the iterative scene state steps (3) and (3) are adoptedEquation (9) predicts the time t s +1 pedestrian coordinate at time t ob +1 as the observation start time, at time t s +1 as the observation end time, the time t being predicted by this method s +2 pedestrian coordinates, continuing to adopt the method to predict until the predicted time t s +t pred Pedestrian coordinates of, t of the embodiment pred Is 12, a prediction model is constructed.
The other steps were the same as in example 1.
Claims (3)
1. A pedestrian trajectory prediction method based on a gate cycle unit network of scene state iteration is characterized by comprising the following steps:
(1) Data pre-processing
Extracting pedestrian trajectory data from the public data set ETH and the data set UCY, including 5 sub-data sets and the nonlinear trajectory of the pedestrian, wherein the two-dimensional coordinates of each pedestrian i at time t are extracted from the data set ETH and the data set UCY and recorded as
All the pedestrian coordinate data are processed by a coordinate standardization and data enhancement method;
(2) Extracting scene states
1) Respectively coding each pedestrian i at a video moment t by using an independent gate cycle unit network, coding all the pedestrians i at the moment t by using the same method, inputting the current position information of the pedestrian i and the hidden layer state of the gate cycle unit network into the gate cycle unit network, and adopting tensor
The position information of the pedestrian i at the time t is determined according to the formula (1):
where phi (-) is a nonlinear function with an offset ReLU, W e Is a weight matrix of the function, b e Is a bias matrix of the function;
2) Determining the hidden layer state of the pedestrian i at the time t according to the formula (2):
wherein, the first and the second end of the pipe are connected with each other,
is the hidden layer state W of the pedestrian i at the time t-1 in the gate cycle unit network G Is an internal weight matrix of the gate-cycle cell network input, b G An internal bias matrix that is a network input to the gate cycle unit;
3) Determining the spatial relative position relationship between pedestrians in a scene according to the formula (3)
Wherein phi l Is a ReLU nonlinear function, W l Is the weight of the function and is,
and &>
Respectively representing the space coordinates of a pedestrian i and a pedestrian j at a moment t, wherein i and j are limited positive integers, and i is not equal to j;
4) Determining feature matrix of relation between pedestrians according to equations (4) and (5)
Wherein sigma F Is a nonlinear function with offset Sigmoid, W F Is a weight matrix of the function, b F Is a bias matrix of the function;
5) Determining the attention of a pedestrian in a scene according to equation (6)
Wherein W α Is a weight matrix of the Softmax function;
(3) Iterative scene states
Iterating the states of the hidden layers of the pedestrians obtained by the formula (2) according to the formulas (7) and (8) to obtain updated states
Wherein, denotes a Hadamard product, N is the number of all pedestrians appearing in the scene, is a finite positive integer, W h For the coefficient matrix, z represents the update gate in the gate rotation unit, σ z Is a Sigmoid nonlinear function in the gate, W z Is the weight moment of the functionArray, b z Is a bias matrix of the function;
(4) Building a prediction model
The predicted coordinates of the pedestrian i at the time t +1 are determined according to equation (9)
Wherein W represents a fully learned parameter matrix;
combining the data set ETH with the coordinate series provided by data set UCY, at time t ob To observe the starting moment, the moment t s For observing the end time, determining a complete observation step length period P:
P＝t s -t ob (10)
all information in the scene state in the step length period P is transmitted to a gate cycle unit network, and the step (3) and the formula (9) of the iterative scene state are adopted to predict the time t s +1 pedestrian coordinate at time t ob +1 as the observation start time, at time t s +1 as the observation end time, the time t being predicted by this method s +2 pedestrian coordinates, continuing to adopt the method to predict until the predicted time t s +t pred The pedestrian coordinates of the vehicle are constructed into a prediction model;
(5) Pedestrian trajectory prediction
Repeatedly training the prediction model by using a leave-one-out method for the data set ETH and the data set UCY, minimizing the mean square error MSE, and determining the mean square error MSE of the predicted track and the real track according to the formula (11):
and obtaining the optimal training model parameters, applying the optimal training model parameters to the training model, inputting coordinate data to be predicted, and predicting the pedestrian track.
2. The method for predicting pedestrian trajectories based on the gate loop unit network of the scene state iteration as claimed in claim 1, wherein in the step (1) of data preprocessing, the coordinate normalization method comprises: within the observation time length, taking the initial coordinate of the pedestrian i as an origin; the data enhancement method comprises the following steps: and randomly rotating the video image of the corresponding frame.
3. The iterative gate-loop unit network pedestrian trajectory prediction method based on scene states of claim 1, characterized in that: in the step (4) of constructing the prediction model, the value range of the observation step period P is as follows: p is as [5,10 ]](ii) a Said t pred The value range of (A) is as follows: t is t pred ∈[8，12]。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010319857.5A CN111553232B (en) | 2020-04-22 | 2020-04-22 | Gate loop unit network pedestrian trajectory prediction method based on scene state iteration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010319857.5A CN111553232B (en) | 2020-04-22 | 2020-04-22 | Gate loop unit network pedestrian trajectory prediction method based on scene state iteration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111553232A CN111553232A (en) | 2020-08-18 |
| CN111553232B true CN111553232B (en) | 2023-04-07 |
Family
ID=72005836
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010319857.5A Active CN111553232B (en) | 2020-04-22 | 2020-04-22 | Gate loop unit network pedestrian trajectory prediction method based on scene state iteration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111553232B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112558185A (en) * | 2020-11-19 | 2021-03-26 | 中国石油大学(华东) | Bidirectional GRU typhoon track intelligent prediction and forecast system based on attention mechanism, computer equipment and storage medium |
| 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 |
| CN113689470B (en) * | 2021-09-02 | 2023-08-11 | 重庆大学 | Pedestrian motion trail prediction method under multi-scene fusion |
| CN114446046A (en) * | 2021-12-20 | 2022-05-06 | 上海智能网联汽车技术中心有限公司 | LSTM model-based weak traffic participant track prediction method |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886060A (en) * | 2017-11-01 | 2018-04-06 | 西安交通大学 | Pedestrian's automatic detection and tracking based on video |
| CN108564118B (en) * | 2018-03-30 | 2021-05-11 | 陕西师范大学 | Crowd scene pedestrian trajectory prediction method based on social affinity long-term and short-term memory network model |
| CN109241871A (en) * | 2018-08-16 | 2019-01-18 | 北京此时此地信息科技有限公司 | A kind of public domain stream of people's tracking based on video data |
| CN110737968B (en) * | 2019-09-11 | 2021-03-16 | 北京航空航天大学 | Crowd trajectory prediction method and system based on deep convolutional long and short memory network |
| CN110781838B (en) * | 2019-10-28 | 2023-05-26 | 大连海事大学 | Multi-mode track prediction method for pedestrians in complex scene |
-
2020
- 2020-04-22 CN CN202010319857.5A patent/CN111553232B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111553232A (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553232B (en) | Gate loop unit network pedestrian trajectory prediction method based on scene state iteration | |
| EP4198820A1 (en) | Training method for semi-supervised learning model, image processing method, and device | |
| CN111461258B (en) | Remote sensing image scene classification method of coupling convolution neural network and graph convolution network | |
| CN108256562B (en) | Salient target detection method and system based on weak supervision time-space cascade neural network | |
| US11663441B2 (en) | Action selection neural network training using imitation learning in latent space | |
| Blum et al. | Fishyscapes: A benchmark for safe semantic segmentation in autonomous driving | |
| CN107330410B (en) | Anomaly detection method based on deep learning in complex environment | |
| Hu et al. | Probabilistic future prediction for video scene understanding | |
| Ondruska et al. | Deep tracking: Seeing beyond seeing using recurrent neural networks | |
| Chen et al. | Efficient movement representation by embedding dynamic movement primitives in deep autoencoders | |
| CN112132149B (en) | Semantic segmentation method and device for remote sensing image | |
| CN110084201B (en) | Human body action recognition method based on convolutional neural network of specific target tracking in monitoring scene | |
| CN112771542A (en) | Learning-enhanced neural network based on learned visual entities | |
| CN110570035B (en) | People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
| CN112183742B (en) | Neural network hybrid quantization method based on progressive quantization and Hessian information | |
| Zhang et al. | A multi-modal states based vehicle descriptor and dilated convolutional social pooling for vehicle trajectory prediction | |
| US11789466B2 (en) | Event camera based navigation control | |
| CN115438856A (en) | Pedestrian trajectory prediction method based on space-time interaction characteristics and end point information | |
| CN116434033A (en) | Cross-modal contrast learning method and system for RGB-D image dense prediction task | |
| CN113657387A (en) | Semi-supervised three-dimensional point cloud semantic segmentation method based on neural network | |
| Darapaneni et al. | Autonomous car driving using deep learning | |
| Salem et al. | Semantic image inpainting using self-learning encoder-decoder and adversarial loss | |
| CN115346207A (en) | Method for detecting three-dimensional target in two-dimensional image based on example structure correlation | |
| Robert | The Role of Deep Learning in Computer Vision | |
| Xiao et al. | Fine-grained road scene understanding from aerial images based on semisupervised semantic segmentation networks |
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 |