CN113159403B - Intersection pedestrian track prediction method and device - Google Patents
Intersection pedestrian track prediction method and device Download PDFInfo
- Publication number
- CN113159403B CN113159403B CN202110387223.8A CN202110387223A CN113159403B CN 113159403 B CN113159403 B CN 113159403B CN 202110387223 A CN202110387223 A CN 202110387223A CN 113159403 B CN113159403 B CN 113159403B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- intersection
- preset
- track
- pedestrian track
- 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
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000013528 artificial neural network Methods 0.000 claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 50
- 230000006870 function Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 21
- 238000010606 normalization Methods 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000015654 memory Effects 0.000 claims description 6
- 238000011425 standardization method Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 239000010410 layer Substances 0.000 description 28
- 230000008569 process Effects 0.000 description 24
- 238000004364 calculation method Methods 0.000 description 19
- 230000000306 recurrent effect Effects 0.000 description 19
- 230000001364 causal effect Effects 0.000 description 10
- 238000011156 evaluation Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000004927 fusion Effects 0.000 description 6
- 230000007246 mechanism Effects 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000007774 longterm Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000002759 z-score normalization Methods 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000007480 spreading Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 230000003997 social interaction Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The application discloses a method and a device for predicting pedestrian trajectories at intersections, wherein the method comprises the steps of obtaining pedestrian trajectory data at preset intersections within preset time periods, wherein the pedestrian trajectory data are longitude and latitude position data, and the preset intersections are intersections of at least two paths; the pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection; the method comprises the steps of inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network. The method and the device solve the problem that the existing pedestrian track prediction mode is high in complexity.
Description
Technical Field
The application relates to the field of smart cities, in particular to a method and a device for predicting pedestrian trajectories at intersections.
Background
Pedestrian track prediction is an important research field of smart cities, and is beneficial to urban travel planning, traffic jam relief, urban business planning and other aspects. Pedestrian trajectories are largely divided into long-term trajectories of Location-based social networks (LBSN) and short-term trajectories of consecutive locations. The LBSN is a network established by intelligent terminal equipment and used for social interaction, and is an online platform for users to share information such as interests, hobbies, states and activities. The LBSN provides a location-based service for users that allows users to share their respective locations and details of the locations in a social network. The long-term track prediction task of the pedestrians is to mine potential modes and rules of the data based on the sign-in data of the pedestrians and combined with implicit factors such as interests, hobbies and the like, so as to predict the possible situations of the future pedestrian positions. Short-term trajectory prediction studies primarily perform a series of spatiotemporal feature calculations around a pedestrian's continuous positional changes over a short period of time. The track of the vehicle in the road is limited by the road distribution and traffic signs, and the track route is strong in regularity, so that the model can calculate the space-time correlation more conveniently. However, in real life, the trajectory of pedestrians is complex and variable, subject to static obstacles around and in the vicinity of the pedestrians, and less restricted by traffic regulations, which presents a greater challenge to short-term trajectory prediction of pedestrians.
Currently, there are prediction methods based on deep move, prediction methods based on LSTPM (consisting of a non-local network for calculating long-term preference characteristics and a modified Recurrent neural network), and prediction methods based on ARNN (data is divided into two types, which are processed by an attribute layer and a Recurrent layer consisting of LSTM, the output of the attribute layer is one of inputs of LSTM), an attribute gan (an attribute mechanism is added before an Encoder-Decoder structure consisting of LSTM), and the like. In actually making predictions of pedestrian trajectories, the inventors have found that trajectory prediction can be regarded as a sequence generation task, i.e. predicting future trajectories based on past positions. Recurrent neural network models are often used to learn the general motion of humans and predict their future trajectories. The simple recurrent neural network cannot deal with the problem (Vanishing gradient problem) that the weight exponentially explodes or disappears along with recursion, and is difficult to capture long-term time association, and the problem can be well solved by combining different long-term and short-term memory networks LSTM. Recurrent neural networks may describe dynamic time behavior because, unlike feedforward neural networks (feedforward neural network) which accept more structured inputs, recurrent neural networks cycle states through their own networks and thus may accept a wider range of time series structured inputs, as shown in fig. 1. However, recurrent neural networks have a fatal problem: each element within the sequence is directly related to all elements that precede the current element, and its complexity increases explosively with the computation process.
In summary, the existing prediction method of the pedestrian track has the problem of high complexity.
Disclosure of Invention
The main purpose of the application is to provide a method and a device for predicting the pedestrian track at the intersection, which solve the problem that the existing pedestrian track prediction mode is high in complexity.
To achieve the above object, according to a first aspect of the present application, there is provided a method for predicting an intersection pedestrian trajectory.
The method for predicting the intersection pedestrian track comprises the following steps:
acquiring pedestrian track data in a preset period, wherein the pedestrian track data are longitude and latitude position data, and the preset intersection is an intersection of at least two paths;
the pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection;
the method comprises the steps of inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network.
Optionally, inputting the pedestrian track sequence into a preset track prediction model, and obtaining the pedestrian track prediction result includes:
performing matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence;
and inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into a time convolution neural network, and obtaining the walking direction of the pedestrian after linearization and normalization function processing.
Optionally, the step of converting the pedestrian track data according to a preset coordinate system and performing standardization processing to obtain a pedestrian track sequence includes:
and converting the pedestrian track data according to a preset coordinate system and processing the pedestrian track data by a Z-Score standardization method to obtain a pedestrian track sequence.
Optionally, the normalization function is a Log Softmax based normalization function.
Optionally, the method further comprises:
acquiring historical pedestrian track data of the preset intersection;
and training a model based on the historical pedestrian track data and the time convolution neural network to obtain the preset track prediction model.
Optionally, the preset intersection is an intersection, the preset coordinate system uses the center of the intersection as the origin of the coordinate system, and the central lines of four paths corresponding to the intersection are used as the X and Y axes.
Optionally, the preset intersection is a non-crossroad, the preset coordinate system uses the center of the intersection as the origin of the coordinate system, and uses the center line of a certain road corresponding to the intersection as the X axis or the Y axis.
To achieve the above object, according to a second aspect of the present application, there is provided another apparatus for intersection pedestrian trajectory prediction.
The device for predicting the pedestrian track at the intersection comprises:
the first acquisition module is used for acquiring pedestrian track data in a preset time period, wherein the pedestrian track data are longitude and latitude position data, and the preset intersection is an intersection of at least two roads;
the preprocessing module is used for converting and standardizing the pedestrian track data according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as the origin of the coordinate system, and the X axis and the Y axis are determined by the center lines of a plurality of roads corresponding to the intersection;
the prediction module is used for inputting the pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to the historical pedestrian track data of the preset intersection and training through a time convolution neural network.
Optionally, the prediction module includes:
the processing unit is used for performing matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence;
the prediction unit is used for inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into the time convolution neural network, and obtaining the walking direction of the pedestrian after linearization and normalization function processing.
Optionally, the preprocessing module is further configured to:
and converting the pedestrian track data according to a preset coordinate system and processing the pedestrian track data by a Z-Score standardization method to obtain a pedestrian track sequence.
Optionally, the normalization function is a Log Softmax based normalization function.
Optionally, the apparatus further includes:
the second acquisition module is used for acquiring historical pedestrian track data of the preset intersection;
and the training module is used for training the model based on the historical pedestrian track data and the time convolution neural network to obtain the preset track prediction model.
Optionally, the preset intersection is an intersection, the preset coordinate system uses the center of the intersection as the origin of the coordinate system, and the central lines of four paths corresponding to the intersection are used as the X and Y axes.
Optionally, the preset intersection is a non-crossroad, the preset coordinate system uses the center of the intersection as the origin of the coordinate system, and uses the center line of a certain road corresponding to the intersection as the X axis or the Y axis.
To achieve the above object, according to a third aspect of the present application, there is provided a computer-readable storage medium storing computer instructions for causing the computer to execute the method for intersection pedestrian trajectory prediction according to any one of the above first aspects.
To achieve the above object, according to a fourth aspect of the present application, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor to cause the at least one processor to perform the method of intersection pedestrian trajectory prediction of any one of the above first aspects.
In the method and the device for predicting the pedestrian track of the intersection, pedestrian track data in a preset period of time are obtained, the pedestrian track data are longitude and latitude position data, and the preset intersection is an intersection of at least two roads; the pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection; the method comprises the steps of inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network. It can be seen that, in the embodiment of the present application, the walking direction of the pedestrian at the intersection is predicted according to the track of the pedestrian, and the existing data regression mode (based on the recurrent neural network mode) is converted into a mode of data regression and classification fusion. Specifically, when the prediction is performed, the preset track prediction model is obtained based on training of a time convolution neural network, and the model integrates implicit factors such as track data, a distance sequence, a time sequence and the like to realize a multi-mode data prediction task, so that the traditional recurrent neural network is replaced, and the complexity of time sequence calculation can be reduced.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, are included to provide a further understanding of the application and to provide a further understanding of the application with regard to the other features, objects and advantages of the application. The drawings of the illustrative embodiments of the present application and their descriptions are for the purpose of illustrating the present application and are not to be construed as unduly limiting the present application. In the drawings:
FIG. 1 is a schematic diagram of the relationship between elements in a sequence of a prior art recurrent neural network;
FIG. 2 is a flow chart of a method for predicting the trajectory of an intersection pedestrian provided according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a preset coordinate system provided according to an embodiment of the present application;
FIG. 4 is a schematic diagram of another preset coordinate system provided in accordance with an embodiment of the present application;
FIG. 5 is a schematic diagram of causal convolution and hole convolution in a TCN architecture;
FIG. 6 is a flow chart of another method for intersection pedestrian trajectory prediction provided in accordance with an embodiment of the present application;
FIG. 7 is a schematic diagram of a distribution of predicted outcomes in a plurality of ways provided in accordance with an embodiment of the present application;
FIG. 8 is a schematic representation of ROC curves for various models provided in accordance with embodiments of the present application;
FIG. 9 is a schematic representation of F1 Score index changes for each model training process provided in accordance with an embodiment of the present application;
FIG. 10 is a schematic illustration of model parameters for each model provided in accordance with an embodiment of the present application;
FIG. 11 is a block diagram of an apparatus for intersection pedestrian trajectory prediction provided in accordance with an embodiment of the present application;
fig. 12 is a block diagram of another apparatus for predicting intersection pedestrian trajectories according to an embodiment of the present application.
Detailed Description
In order to make the present application solution better understood by those skilled in the art, the following description will be made in detail and with reference to the accompanying drawings in the embodiments of the present application, it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the present application described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
Firstly, it should be noted that, the method for predicting the intersection pedestrian track in the embodiment of the present application may preset the walking direction in a very short time in the future according to the track of the pedestrian in the current period, so that the prediction mode is applied in intelligent traffic, for example, early warning may be performed for the pedestrian or the vehicle in advance according to the prediction of the walking direction of the pedestrian, so as to reduce the occurrence of accidents; in addition, the flow of the pedestrian walking direction can be predicted according to the prediction of the pedestrian walking direction, so that the duration of the traffic light is adjusted, and the traffic is more intelligent.
According to an embodiment of the application, a method for predicting the trajectory of a pedestrian at an intersection is provided, as shown in fig. 2, and the method comprises the following steps:
s101, acquiring pedestrian track data of a preset intersection within a preset period.
The preset intersections are intersections of at least two roads, such as T-shaped intersections, three-way intersections, crossroads and the like, and intersections with the traveling directions being multiple directions exist. In particular, the intersections of several paths, embodiments of the present application are not limited. The models corresponding to different preset intersections have some differences, and the main differences are that the selection of the preset coordinate system is different in the data preprocessing stage, and in addition, the differences are described in more detail in the follow-up steps of different prediction results (different types of walking directions).
The embodiment of the application is mainly aimed at the prediction of short-term pedestrian trajectories, so that the preset period is a shorter time period, for example, may be 20s,30s, etc. The preset time period is consistent with the selection of training data during training of the preset track prediction model in the subsequent step. That is, the training sample selects the pedestrian track data in the period of time, so the pedestrian track selected in the process of prediction is also in the period of time.
The pedestrian track data are longitude and latitude position data, and the longitude and latitude data can be obtained through the positioning equipment.
S102, converting the pedestrian track data according to a preset coordinate system and carrying out standardization processing to obtain a pedestrian track sequence.
In practical application, because the accuracy of directly obtained longitude and latitude position data is relatively poor, the prediction of short-term corresponding short-distance pedestrian tracks cannot be satisfied, for example, for two position points with actual distance of 7 meters, the difference can be displayed only by 7 th and 8 th positions after the position points are possibly accurate in the longitude and latitude position data, so that the prediction is inaccurate based on the longitude and latitude position data, and the accurate prediction is more suitable after the tiny difference is properly amplified. Therefore, the acquired longitude and latitude position data needs to be preprocessed.
Specifically, the pretreatment process in this embodiment includes: and converting the pedestrian track data according to a preset coordinate system, and performing standardization. The preset coordinate system takes the center of the intersection as the origin of the coordinate system, and the X axis and the Y axis are determined by the center lines of the multiple roads corresponding to the intersection. It should be noted that, the determination modes of the origins of the coordinate systems corresponding to the preset intersections of different types are the same, and the determination modes of different X-axis and Y-axis are different. Specifically, as shown in fig. 3, if the preset intersection is an intersection, the central lines of the four roads corresponding to the intersection are taken as X and Y axes. As shown in fig. 4, if the preset intersection is a non-intersection, the preset coordinate system uses the center of the intersection as the origin of the coordinate system, and uses the center line of a certain road corresponding to the intersection as the X-axis or the Y-axis.
The method comprises the steps of converting coordinates of a pedestrian track based on a preset coordinate system, and obtaining two sequences { X } and { Y } after conversion, wherein the sequence { X } is an abscissa set of the track in the coordinate system, and the sequence { Y } is an ordinate set of the track in the coordinate system. The normalization is performed after the conversion, and the specific normalization may be a Z-Score normalization method or a normalization method such as Min-Max and Decimal Scaling, but in actual experiments, the Z-Score normalization method is more effective. The Z-Score normalization method is applicable to cases where the maximum and minimum values of the data are unknown, or where there is outlier data outside the range of values. Different from the running mode of the vehicle, the behavior of the person is complex and changeable, and the situation of curve running or finding a road shortcut occurs, so that the Z-Score standardization method is suitable for application scenes of pedestrian track prediction.
S103, inputting the pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of the pedestrian at a preset intersection in a future time period.
The preset track prediction model is a model obtained by obtaining training samples according to historical pedestrian track data of a preset intersection and training in advance through a time convolution neural network. Different types of preset intersections obtain different types of preset track prediction models. In addition, the pedestrian track conditions of different intersections may also have large differences, so preset intersections of the same type but different positions can also correspond to different preset track prediction models. The principle of training of different models is the same, except for training samples.
Specifically, each training sample comprises pedestrian track data in a preset period and the walking direction of pedestrians after the preset period. For the walking direction, as shown in fig. 3, the specific labeling manner given in this embodiment is that the walking direction is determined by the final position of the pedestrian after passing through the road, and different roads can be marked with 1, 2, 3 and 4 respectively. If there are three walking directions, they can be marked as 1, 2, 3. Before the training samples are input into the time convolution neural network for training, the preprocessing process in the step S102 is carried out, a corresponding pedestrian track sequence is obtained after preprocessing, and then matrix conversion, linearization and convolution processing are carried out on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence; and then inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into a time convolution neural network for training, finally obtaining a preset track prediction model which is input as a pedestrian track sequence and output as a pedestrian walking direction mark. The model training process is the same as the subsequent model-based walking direction prediction process, so that reference can be made to the subsequent prediction process, and details are not repeated here.
It should be noted that, in the embodiment of the present application, the conventional recurrent neural network is replaced by a time convolution neural network, and the time convolution neural network uses the idea of large-scale parallel processing of the convolution neural network to map a multidimensional matrix into a time sequence, and obtains a sufficiently large receptive field through a multi-layer network, so as to perform deep network parallel processing. As shown in fig. 5, which is a causal convolution and a hole convolution in the TCN architecture, it can be seen from fig. 5 that knowing the value of each layer depends only on the historical value of the previous layer, which characterizes the causal convolution. The extraction of the information of the previous layer of each layer is jumping, and the layer-by-layer hole factor d grows in an index of 2, so that the characteristic of hole convolution is reflected. The embodiment of the application precisely refers to the calculation modes of spatial correlation and time dependence, integrates various implicit factors, increases the space-time characteristics of data, and reduces the complexity of time sequence calculation.
Specifically, "inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result" includes the following steps:
1) Performing matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence;
The method comprises the steps of carrying out matrix conversion on a pedestrian track sequence, specifically connecting the standardized sequences { X } and { Y } obtained in the previous steps through a concatate to obtain a data matrix, then carrying out linearization to obtain the input dimension of Conv1d, and locally connecting to enable Conv1d to extract local features of data. The convolution process is to make template matching in each local area of the matrix. Pooling of Conv1d is a down-sampling process where the relative relationship between different features plays an important role, which can increase the generalization ability of the model by computing translational relationships and controlling the over-fitting. The pedestrian track characteristic data comprising the characteristics of each node and the correlation characteristics among the nodes in the pedestrian track sequence can be obtained through the matrix conversion, the linear processing and the convolution processing.
2) And inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into a time convolution neural network, and obtaining the walking direction of the pedestrian after linearization and normalization function processing.
The distance auxiliary sequence and the time auxiliary sequence are sequences composed of auxiliary factors Auxiliary Factors, and the paving factors are calculated according to the difference value between the first node and other nodes in the pedestrian track sequence, as shown in formula 1:
A i ＝a i -a 0 ,i∈[0,len(a)]Equation 1
Wherein A represents the sequence of different cofactors, a i Represents the (i+1) th node, a 0 The 1 st node is represented, and len (a) represents the length of the pedestrian trajectory sequence, that is, the number of nodes.
It should be noted that, equation 1 gives a generation manner of the cofactor sequence, and specific calculation manners of different cofactors need adaptive adjustment. For the calculation formula of the distance auxiliary sequence in the application, adjustment is needed, specifically, the distance difference value (distance auxiliary factor) between different nodes can be calculated according to the following formula 2:
wherein d represents the sequence of distance cofactors, d i The ith distance-spreading factor in the sequence representing the distance-spreading factor is the distance difference between the (i+1) th node and the 1 st node, x i And y i Values representing the horizontal and vertical axes of the (i+1) th node, x 0 And y 0 Values representing the horizontal and vertical axes of node 1.
In addition, the spreading aid factors in the application also include time auxiliary factors, and the calculation mode of each time auxiliary factor in a specific time spreading aid sequence can be determined by referring to the calculation mode of the formula 1.
After the distance auxiliary sequence and the time auxiliary sequence are determined according to the mode, the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence are input into a time convolution neural network for training, and the specific time convolution neural network comprises a plurality of Residual blocks which are connected through shorcut connection so that the data are easier to optimize. If the objective function h (x) is approximated by a nonlinear element f (x, θ), the objective function can be split into an identity function x and a residual function h (x) -x, and a nonlinear element made up of a neural network approximates the original objective function with sufficient power according to the general approximation theorem. Thus, the original problem is translated into: let the linear unit f (x, θ) approximate the residual function h (x) -x and approximate h (x) with f (x, θ) +x. The multi-layer residual structure has also been shown to facilitate the non-linearization process of the deep neural network, while the performance of the single-layer residual structure is not ideal. The residual block in a time convolutional neural network mainly comprises two processes: causal convolution and hole convolution.
The formula of causal convolution Causal Convolution is shown in formula 3, where { b } 1 ,b 2 ,...,b t The input sequence, { c } is 1 ,c 2 ,...,c t The hidden layer output sequence, { f 1 ,f 2 ,...,f k And represents a filter. Causal convolution focuses only on historical information and ignores future information, c t Will only be determined by b t Previous data was derived. And the larger K is, the more historical information can be traced, if the original input sequence of the current layer is [0, i ]]I.e. the subscript is from 0 to i, i.e. the sequence of i+1 nodes, the original input sequence of the next layer becomes [0, i+1 ]]I.e. the subscripts from 0 to i+1, i.e. i+2 nodes constitute the sequence.
The formula of the hole convolution Dilated Convolution is shown in formula 4, wherein d represents a hole factor, which varies exponentially with 2 according to the depth of the network, and increasing either d or K increases the receptive field range. The result at time t of each layer can only be calculated by the data from time 0, t, i.e. 0 to time t, which embodies the idea of causal convolution. The result of each moment is that the value is jumped and taken according to the hole factor in the network of the previous layer, which reflects the idea of hole convolution.
After the multi-layer residual structure, linearization is performed, and then normalization is performed, in this embodiment, a Log Softmax function is used, and compared with a traditional Softmax function, the Log Softmax is added with one Log operation based on the Softmax, and the Log operation is specifically shown in formula 5. The method can solve the problems of data overflow and unrerflow, and can accelerate the operation speed and improve the stability of data.
Wherein m is i The probability value, m, of the class of the highest probability in the classification result j The probability of being classified into each class is the sum of the classification results. Specifically, for the case that the preset intersection is an intersection, the classification result (i.e. the final walking direction of each pedestrian track) includes 1, 2, 3, 4, and four types, for a certain pedestrian track, m i M is the probability of being classified into the class with highest probability among 4 classes j For the sum of the probabilities of classifying into each of the 4 classes.
After Log Softmax, an output result can be finally obtained, and for the case that the preset intersection is an intersection, the output result of each pedestrian track can be any one of 1, 2, 3 and 4, and then the walking direction of the pedestrian can be determined according to the corresponding relation between 1, 2, 3 and 4 and the road.
In addition, in the whole prediction process, the future position coordinates are predicted firstly according to the pedestrian track, the future position can be predicted to obtain the future track, and the final walking direction can be determined by the track prediction. The prediction of the position corresponds to an intermediate result, the end result being the direction of travel. The embodiment of the application converts the existing data regression mode into a mode of data regression and classification fusion. The preset track prediction model is a depth space-time model, and the model integrates implicit factors such as track data, distance difference values, time difference values and the like to realize multi-mode data prediction tasks. Wherein data space correlation characteristic calculation (pedestrian track characteristic data corresponding to each node characteristic and correlation characteristics between nodes obtained by convolution processing in the foregoing) is performed using a convolutional neural network, and data time dependence calculation is performed using a time convolutional neural network TCN.
From the above description, it can be seen that in the method for predicting the pedestrian track at the intersection according to the embodiment of the present application, the preset intersection is obtained, the pedestrian track data in the preset period is longitude and latitude position data, and the preset intersection is an intersection of at least two roads; the pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection; the method comprises the steps of inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network. It can be seen that, in the embodiment of the present application, the walking direction of the pedestrian at the intersection is predicted according to the track of the pedestrian, and the existing data regression mode (recurrent neural network) is converted into a mode of data regression and classification fusion. Specifically, when the prediction is performed, the preset track prediction model is obtained based on training of a time convolution neural network, and the model integrates implicit factors such as track data, a distance sequence, a time sequence and the like to realize a multi-mode data prediction task, so that the traditional recurrent neural network is replaced, and the complexity of time sequence calculation can be reduced.
Further, another method for predicting a pedestrian track at an intersection is provided in the embodiment of the present application, as shown in fig. 6, where the intersection is illustrated, input data in fig. 6 is longitude and latitude position data of a pedestrian track in a preset period obtained originally, the pedestrian track Sequence trajectory Sequence is obtained after processing, and then a Spatial Convh process is performed, including matrix conversion (concatenate Sequence), linearization (Linearization) and convolution (Conv 1 d) of the pedestrian track Sequence to obtain pedestrian track feature data including features of each node in the pedestrian track Sequence and correlation features between nodes, and then the pedestrian track feature data is simultaneously input into a Time convolution neural network (technical Conv) with two auxiliary sequences (Distance Sequence, time Sequence), where the technical Conv includes a multi-layer residual structure, and after output from the technical Conv, the pedestrian track Sequence is subjected to linear society and Log society, and finally a walking direction prediction result, i.e. "1, 2, 3, 4" and a walking direction corresponding to each intersection is obtained.
Further, as a supplement or refinement of the above embodiment, the following is included:
In order to verify the effect of the prediction mode of the present application, comparative analysis was performed with other various existing modes, specifically as follows:
the prediction mode and the existing modes based on other models in the application are subjected to performance comparison through four evaluation indexes. The specific four evaluation indexes are Accuracy, precision, recall, F1 Score. The Accuracy is an evaluation index of the traditional classification problem and represents the percentage of the correct result of model prediction to the total sample; precision represents the probability of actually positive samples among all predicted positive samples; recall represents the probability of being predicted as a positive sample among samples that are actually positive; the Precision and Recall indicators are sometimes offset by the fact that the Precision is high and the Recall is reduced, and in some scenarios both Precision and Recall are considered, the most common approach is F1 Score. The following table shows the calculation modes of four evaluation indexes:
| Accuracy | (TP+TN)/(TP+TN+FP+FN) |
| Precision | TP/(TP+FP) |
| Recall | TP/(TP+FN) |
| F1 Score | (2×Precision×Recall)/(Precision+Recall) |
the model predicts the positive classes as the number of the positive classes; FN, predicting positive classes as the number of negative classes by the model; FP, model predicts negative class as the number of positive class; TN the model predicts the negative classes as the number of negative classes.
After the evaluation index is determined, the same pedestrian trajectory data is respectively predicted based on 3 commonly used recurrent neural networks (RNN, LSTM, GRU) and 5 existing pedestrian prediction special models (the structures of the five models are as follows), and the results are shown in the following table.
Structure of five models:
Fuzzy-LSTM the model sets two LSTM modules to calculate the periodicity and proximity of the tracks, respectively. The two LSTM module outputs are ultimately merge.
The main idea of the model is based on recurrent neural networks, which encodes an input sequence into a fixed length vector, and then decodes the vector into an output sequence.
AttenGAN-the model was built up of an LSTM composed of an Encoder-Decoder structure preceded by an Attention mechanism.
ARNN data is divided into two types, namely, the data is processed by an attribute layer and a current layer composed of LSTM. The output of the Attention layer is one of the inputs of the LSTM.
DeepMove track data first passes through multi-modal embedding layer, and then computes track features through GRU and Attention mechanisms.
From the results in the above table, it can be seen that the approach of the present application has the highest accuracy compared to the existing model, and performs best in comparison of the four evaluation indexes. Compared with the common recurrent neural network (RNN, LSTM, GRU), the evaluation index is improved by 84.08 percent at maximum. Compared with the combination of the spatial correlation and the time dependence, the comparison result of the method and the recurrent neural network proves that the performance of the model can be obviously improved by only performing the calculation mode of the time dependence. The evaluation index is improved by 29.86% at maximum compared with a special deep learning model (Fuzzy-LSTM, encoder-Decoder, attenGAN, deepMove). The comparison result of the mode of the application and other models with special structures proves the performance advantage of the combination of the convolutional neural network and the TCN in track prediction. In the method, the accuracy of the model can be improved by increasing the network layer number of the TCN. However, increasing the number of layers sacrifices the speed of training. In the practical application scenario, the ratio of the training speed to the accuracy of the model should be comprehensively considered.
In addition, several kinds of prediction results of better performing models are compared through the distribution situation of the prediction results, specifically, as shown in fig. 7, where a represents ARNN, b represents deep, and c represents the mode of the present application). From the experimental results, we can understand that the three models have better prediction results for the fourth category, because the fourth category has more data volume than the other categories in the actual data set, and the sample features are rich. The present approach shows good accuracy even in the case of less data of the first category. This is because the method of the present application makes full use of the history information of the track, all of which participate in the calculation process of temporal convolution layer, and this process fully exploits the position information from the pedestrian start point to the intersection center point.
In addition, in order to more intuitively demonstrate the accuracy of the method, the ROC curve is used to illustrate that the closer the ROC curve is to the upper left corner, the higher the recall of the model, i.e., the point on the ROC curve is the least classification error. To compare the performance of different models, ROC curves for the respective models are plotted into the same coordinate system, as shown in fig. 8. The mode, deep move and ARNN have obvious precision advantages, and the space-time model structure enables the space correlation and time dependence to be reserved. The method uses dynamically-changing receptive fields and longer-term historical information, so that the method can more accurately capture the space-time relationship and the recursion relationship among all nodes of the pedestrian track.
As described above, the cavity convolution mechanism and the pure convolution structure of TCN strongly promote the model convergence speed of the training process. The existing special models are mostly designed based on traditional recursive networks, and they cannot avoid the single-step calculation mode of the recursive network. The model and the F1 Score index change of the model training process in the mode of the application are shown in fig. 9, and from fig. 9, it can be known that the performance change of AttenGAN, ARNN and deep move in the early stage of model training is slower, and the performance of the model reaches the optimal performance only in the ending stage of the whole training process. The strong model convergence capability in the method enables the model to reach the optimal performance value in the early stage of training, and in the whole model training stage, the model does not have the fitting phenomenon. The experimental results of the 3-layer structure and the 5-layer structure of the application mode prove that in a certain range, the performance and the speed of the model in the application mode can be improved by increasing the number of TCN network layers, and high robustness can not be influenced.
Because the deep learning relies on network parameters of tens of millions in the neural network to participate in calculation together, the defects of complex network structure, large operation amount and low speed exist, and the deep learning is difficult to be transplanted into embedded equipment. As the number of network model layers gets deeper, parameters get more and more, and it is important to reduce their size and computational loss. The causal convolution mechanism in the application mode realizes the tracing and fusion of the historical information at the characteristic level, and unlike deep move, the causal convolution mechanism only carries out the calculation process on the historical information at the data level, and the process leads deep move to have huge parameter quantity. The model in this embodiment is a lightweight complete convolution model. The model parameters are the smallest compared to other special pedestrian prediction models, as shown in fig. 10.
Finally, summarizing the beneficial effects of the intersection pedestrian track prediction method:
1. converting the existing data regression mode into a data regression and classification fusion mode;
2. the depth space-time model is provided, and the model integrates implicit factors such as track data, distance difference values, time difference values and the like to realize a multi-mode data prediction task, so that the complexity of time sequence calculation is reduced.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
According to an embodiment of the present application, there is further provided an apparatus for implementing the above-mentioned method embodiment for predicting a path pedestrian trajectory, as shown in fig. 11, where the apparatus includes:
the first obtaining module 21 is configured to obtain pedestrian track data in a preset period, where the pedestrian track data is longitude and latitude position data, and the preset intersection is an intersection of at least two paths;
the preprocessing module 22 is configured to convert and normalize the pedestrian track data according to a preset coordinate system to obtain a pedestrian track sequence, where the preset coordinate system uses a center of an intersection as an origin of the coordinate system, and determines an X axis and a Y axis by using center lines of multiple paths corresponding to the intersection;
The prediction module 23 is configured to input the pedestrian track sequence into a preset track prediction model, to obtain a pedestrian track prediction result, where the prediction result is a walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training the training sample through a time convolution neural network.
From the above description, it can be seen that, in the device for predicting the pedestrian track at the intersection according to the embodiment of the present application, the preset intersection is obtained, the pedestrian track data in the preset period is longitude and latitude position data, and the preset intersection is an intersection of at least two paths; the pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection; the method comprises the steps of inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network. It can be seen that, in the embodiment of the present application, the walking direction of the pedestrian at the intersection is predicted according to the track of the pedestrian, and the existing data regression mode (recurrent neural network) is converted into a mode of data regression and classification fusion. Specifically, when the prediction is performed, the preset track prediction model is obtained based on training of a time convolution neural network, and the model integrates implicit factors such as track data, a distance sequence, a time sequence and the like to realize a multi-mode data prediction task, so that the traditional recurrent neural network is replaced, and the complexity of time sequence calculation can be reduced.
Further, as shown in fig. 12, the prediction module 23 includes:
a processing unit 231, configured to perform matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data including each node feature and correlation features between nodes in the pedestrian track sequence;
the prediction unit 232 is configured to input the pedestrian trajectory feature data, the distance auxiliary sequence and the time auxiliary sequence into the time convolution neural network, and obtain a walking direction of the pedestrian after processing by using linearization and normalization functions.
Further, the preprocessing module 22 is further configured to:
and converting the pedestrian track data according to a preset coordinate system and processing the pedestrian track data by a Z-Score standardization method to obtain a pedestrian track sequence.
Further, the normalization function is a Log Softmax based normalization function.
Further, as shown in fig. 12, the apparatus further includes:
a second obtaining module 24, configured to obtain historical pedestrian track data of the preset intersection;
and the training module 25 is used for training the model based on the historical pedestrian track data and the time convolution neural network to obtain the preset track prediction model.
Further, the preset intersection is an intersection, the preset coordinate system takes the center of the intersection as the origin of the coordinate system, and the central lines of four paths corresponding to the intersection are taken as X and Y axes.
Further, the preset intersection is a non-crossroad, the preset coordinate system takes the center of the intersection as the origin of the coordinate system, and the center line of a certain road corresponding to the intersection is taken as the X axis or the Y axis.
Specifically, the specific process of implementing the functions of each unit and module in the apparatus of the embodiment of the present application may refer to the related description in the method embodiment, which is not repeated herein.
According to an embodiment of the present application, there is further provided a computer readable storage medium, where the computer readable storage medium stores computer instructions for causing the computer to execute the method for predicting the intersection pedestrian trajectory in the above method embodiment.
According to an embodiment of the present application, there is also provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executable by the at least one processor to cause the at least one processor to perform the method of intersection pedestrian trajectory prediction in the method embodiment described above.
It will be apparent to those skilled in the art that the modules or steps of the application described above may be implemented in a general purpose computing device, they may be centralized on a single computing device, or distributed across a network of computing devices, or they may alternatively be implemented in program code executable by computing devices, such that they may be stored in a memory device and executed by computing devices, or individually fabricated as individual integrated circuit modules, or multiple modules or steps within them may be fabricated as a single integrated circuit module. Thus, the present application is not limited to any specific combination of hardware and software.
The foregoing description is only of the preferred embodiments of the present application and is not intended to limit the same, but rather, various modifications and variations may be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims (9)
1. A method of intersection pedestrian trajectory prediction, the method comprising:
acquiring pedestrian track data in a preset period, wherein the pedestrian track data are longitude and latitude position data, and the preset intersection is an intersection of at least two paths;
The pedestrian track data are converted and standardized according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as an origin of the coordinate system, and an X axis and a Y axis are determined by the center lines of a plurality of paths corresponding to the intersection;
inputting a pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to historical pedestrian track data of the preset intersection and training through a time convolution neural network;
the step of inputting the pedestrian track sequence into a preset track prediction model, and the step of obtaining a pedestrian track prediction result comprises the following steps:
performing matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence;
and inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into a time convolution neural network, and obtaining the walking direction of the pedestrian after linearization and normalization function processing.
2. The method for predicting the pedestrian track at the intersection according to claim 1, wherein the step of converting and normalizing the pedestrian track data according to a preset coordinate system to obtain the pedestrian track sequence comprises the steps of:
and converting the pedestrian track data according to a preset coordinate system and processing the pedestrian track data by a Z-Score standardization method to obtain a pedestrian track sequence.
3. The method of intersection pedestrian trajectory prediction of claim 1, wherein the normalization function is a Log Softmax-based normalization function.
4. The method of intersection pedestrian trajectory prediction of claim 1, further comprising:
acquiring historical pedestrian track data of the preset intersection;
and training a model based on the historical pedestrian track data and the time convolution neural network to obtain the preset track prediction model.
5. The method for predicting the trajectory of a pedestrian at an intersection according to claim 1, wherein the preset intersection is an intersection, the preset coordinate system is an origin of the coordinate system with a center of the intersection as an origin, and the centerlines of four roads corresponding to the intersection are X and Y axes.
6. The method for predicting the trajectory of a pedestrian at an intersection according to claim 1, wherein the preset intersection is a non-intersection, the preset coordinate system is an origin of the coordinate system with a center of the intersection as an X-axis or a Y-axis, and a center line of a certain road corresponding to the intersection.
7. An apparatus for predicting the trajectory of a pedestrian at an intersection, the apparatus comprising:
the first acquisition module is used for acquiring pedestrian track data in a preset time period, wherein the pedestrian track data are longitude and latitude position data, and the preset intersection is an intersection of at least two roads;
the preprocessing module is used for converting and standardizing the pedestrian track data according to a preset coordinate system to obtain a pedestrian track sequence, wherein the preset coordinate system takes the center of an intersection as the origin of the coordinate system, and the X axis and the Y axis are determined by the center lines of a plurality of roads corresponding to the intersection;
the prediction module is used for inputting the pedestrian track sequence into a preset track prediction model to obtain a pedestrian track prediction result, wherein the prediction result is the walking direction of a pedestrian at a preset intersection in a future time period, and the preset track prediction model is a model obtained by obtaining a training sample according to the historical pedestrian track data of the preset intersection and training through a time convolution neural network;
the step of inputting the pedestrian track sequence into a preset track prediction model, and the step of obtaining a pedestrian track prediction result comprises the following steps:
performing matrix conversion, linearization and convolution on the pedestrian track sequence to obtain pedestrian track feature data containing each node feature and correlation features among nodes in the pedestrian track sequence;
And inputting the pedestrian track characteristic data, the distance auxiliary sequence and the time auxiliary sequence into a time convolution neural network, and obtaining the walking direction of the pedestrian after linearization and normalization function processing.
8. A computer-readable storage medium storing computer instructions for causing the computer to perform the method of intersection pedestrian trajectory prediction of any one of claims 1 to 6.
9. An electronic device, comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor to cause the at least one processor to perform the method of intersection pedestrian trajectory prediction of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110387223.8A CN113159403B (en) | 2021-04-13 | 2021-04-13 | Intersection pedestrian track prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110387223.8A CN113159403B (en) | 2021-04-13 | 2021-04-13 | Intersection pedestrian track prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113159403A CN113159403A (en) | 2021-07-23 |
| CN113159403B true CN113159403B (en) | 2024-03-12 |
Family
ID=76889830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110387223.8A Active CN113159403B (en) | 2021-04-13 | 2021-04-13 | Intersection pedestrian track prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159403B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113569980B (en) * | 2021-08-12 | 2023-09-01 | 中山大学 | Pedestrian movement track online prediction method and system in complex environment |
| CN114338974A (en) * | 2021-12-02 | 2022-04-12 | 深圳市领航卫士安全技术有限公司 | Multi-channel activity path determination method, device, equipment and storage medium |
| CN116560377A (en) * | 2023-05-31 | 2023-08-08 | 北京百度网讯科技有限公司 | Automatic driving model for predicting position track and training method thereof |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101951595B1 (en) * | 2018-05-18 | 2019-02-22 | 한양대학교 산학협력단 | Vehicle trajectory prediction system and method based on modular recurrent neural network architecture |
| CN111091708A (en) * | 2019-12-13 | 2020-05-01 | 中国科学院深圳先进技术研究院 | Vehicle track prediction method and device |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| CN111222438A (en) * | 2019-12-31 | 2020-06-02 | 的卢技术有限公司 | Pedestrian trajectory prediction method and system based on deep learning |
| WO2020164089A1 (en) * | 2019-02-15 | 2020-08-20 | Bayerische Motoren Werke Aktiengesellschaft | Trajectory prediction using deep learning multiple predictor fusion and bayesian optimization |
| CN112215423A (en) * | 2020-10-13 | 2021-01-12 | 西安交通大学 | Pedestrian trajectory prediction method and system based on trend guiding and sparse interaction |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583151B (en) * | 2019-02-20 | 2023-07-21 | 阿波罗智能技术(北京)有限公司 | Method and device for predicting running track of vehicle |
-
2021
- 2021-04-13 CN CN202110387223.8A patent/CN113159403B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101951595B1 (en) * | 2018-05-18 | 2019-02-22 | 한양대학교 산학협력단 | Vehicle trajectory prediction system and method based on modular recurrent neural network architecture |
| WO2020164089A1 (en) * | 2019-02-15 | 2020-08-20 | Bayerische Motoren Werke Aktiengesellschaft | Trajectory prediction using deep learning multiple predictor fusion and bayesian optimization |
| CN111091708A (en) * | 2019-12-13 | 2020-05-01 | 中国科学院深圳先进技术研究院 | Vehicle track prediction method and device |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| CN111222438A (en) * | 2019-12-31 | 2020-06-02 | 的卢技术有限公司 | Pedestrian trajectory prediction method and system based on deep learning |
| CN112215423A (en) * | 2020-10-13 | 2021-01-12 | 西安交通大学 | Pedestrian trajectory prediction method and system based on trend guiding and sparse interaction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113159403A (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11657708B2 (en) | Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM | |
| CN113159403B (en) | Intersection pedestrian track prediction method and device | |
| Akhtar et al. | A review of traffic congestion prediction using artificial intelligence | |
| CN104462190B (en) | A kind of online position predicting method excavated based on magnanimity space tracking | |
| CN108986453A (en) | A kind of traffic movement prediction method based on contextual information, system and device | |
| CN114330868A (en) | Passenger flow prediction method based on self-attention personalized enhanced graph convolution network | |
| CN111582559A (en) | Method and device for estimating arrival time | |
| CN110570035A (en) | people flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
| CN116307152A (en) | Traffic prediction method for space-time interactive dynamic graph attention network | |
| CN114492978A (en) | Time-space sequence prediction method and device based on multi-layer attention mechanism | |
| Liu et al. | Deep learning enabled vehicle trajectory map‐matching method with advanced spatial–temporal analysis | |
| Xing et al. | A data fusion powered bi-directional long short term memory model for predicting multi-lane short term traffic flow | |
| Kawasaki et al. | Multimodal trajectory predictions for autonomous driving without a detailed prior map | |
| CN111722306A (en) | Typhoon landing intensity prediction method and system based on TCN network model | |
| Lv et al. | A deep spatial-temporal network for vehicle trajectory prediction | |
| Xiao et al. | Parking prediction in smart cities: A survey | |
| CN114202120A (en) | Urban traffic travel time prediction method aiming at multi-source heterogeneous data | |
| Jabbar et al. | Smart Urban Computing Applications | |
| Zheng et al. | A deep learning–based approach for moving vehicle counting and short-term traffic prediction from video images | |
| Zhang et al. | Bus passenger flow statistics algorithm based on deep learning | |
| CN116665448A (en) | Traffic speed real-time prediction method and system based on graph convolution network | |
| Joseph et al. | Investigation of deep learning methodologies in intelligent green transportation system | |
| CN115600833A (en) | Smart city restriction scheme determination method, internet of things system, device and medium | |
| CN114418606A (en) | Network taxi appointment order demand prediction method based on space-time convolutional network | |
| CN113962460A (en) | Urban fine-grained flow prediction method and system based on space-time contrast self-supervision |
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 |