CN116259176B - Pedestrian track prediction method based on intention randomness influence strategy - Google Patents
Pedestrian track prediction method based on intention randomness influence strategy Download PDFInfo
- Publication number
- CN116259176B CN116259176B CN202310130465.8A CN202310130465A CN116259176B CN 116259176 B CN116259176 B CN 116259176B CN 202310130465 A CN202310130465 A CN 202310130465A CN 116259176 B CN116259176 B CN 116259176B
- Authority
- CN
- China
- Prior art keywords
- intention
- pedestrian
- vector
- random
- 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 53
- 239000013598 vector Substances 0.000 claims abstract description 105
- 230000008859 change Effects 0.000 claims abstract description 50
- 238000006073 displacement reaction Methods 0.000 claims abstract description 35
- 238000009826 distribution Methods 0.000 claims abstract description 8
- 238000010276 construction Methods 0.000 claims abstract description 7
- 238000005070 sampling Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 7
- 238000009827 uniform distribution Methods 0.000 claims description 3
- 238000007792 addition Methods 0.000 claims 1
- 230000033001 locomotion Effects 0.000 abstract description 7
- 230000006870 function Effects 0.000 description 10
- 206010063385 Intellectualisation Diseases 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 206010039203 Road traffic accident Diseases 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000034994 death Effects 0.000 description 2
- 231100000517 death Toxicity 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007430 reference method Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- 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
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a pedestrian track prediction method based on an intention randomness influence strategy, which relates to the technical field of track prediction and comprises the steps of extracting displacement information of a last frame from a pedestrian history track; obtaining a random number factor from standard normal distribution sampling; inputting displacement information of the last frame into a linear layer with offset, and multiplying the obtained result by a random number factor to obtain a random intention vector for simulating the influence of pedestrian motion randomness; performing second-order fitting on each pedestrian history track coordinate point to obtain a fitting error, initializing the fitting error, and inputting the initialized fitting error into a linear layer with random bias to obtain an intention change probability vector of each pedestrian; the random intention vector and the intention change probability vector are input into a sparse intention module, the random intention vector used by pedestrians or a zero vector which does not bring any influence is determined, an output vector is obtained, and the construction of the random intention module is completed.
Description
Technical Field
The invention relates to the technical field of track prediction, in particular to a pedestrian track prediction method based on an intention randomness influence strategy.
Background
In urban traffic environment, how to find out pedestrians entering a traffic lane and recognize their movement intention and movement track early is a difficult problem to improve the safety of vehicles. According to the world health organization '2018 global road safety status report', the number of deaths due to traffic accidents has risen by 135 ten thousand per year, and on average, one person loses life on the road every 24 seconds, with about 23% of the deaths being pedestrians. The report also emphasizes that road traffic injuries are now the primary killer for children and young adults 5 to 29 years old. How to improve the running safety of vehicles and reduce the accident occurrence number of road traffic is one of the important problems to be solved at present, and only if the moving direction of pedestrians is accurately predicted, potential pedestrians which can possibly happen accidents are avoided in time, the probability of collision of the pedestrians and the vehicles can be reduced.
The influence on the advancing direction and the speed caused by the change of the intention of the pedestrians is seldom considered in the current method, on one hand, the randomness of the motion of the pedestrians is difficult to describe, and on the other hand, the pedestrians under the social rules are limited by groups, so that the current method focuses on the interaction of the pedestrians and the physical environment. However, the randomness of pedestrians is not ignored, and particularly in complex scenes, the high-level unmanned ground application often has to avoid accidents, particularly the prediction of the pedestrian track in traffic scenes is critical to traffic safety; if the track change caused by the randomness of the pedestrian movement can be modeled, the automatic driving vehicle can also run more safely, so that the automatic driving vehicle is widely applied to cities.
Disclosure of Invention
In view of the above problems, the present invention provides a pedestrian trajectory prediction method based on an intention randomness impact strategy, comprising the following steps:
extracting displacement information of the last frame from the historical track of the pedestrian, wherein the displacement information comprises the current speed direction and the current speed of the pedestrian;
obtaining a random number factor from standard normal distribution sampling for simulating the change amount of pedestrian speed;
inputting the displacement information of the last frame into a linear layer with offset, and multiplying the obtained result by a random number factor to obtain a random intention vector;
based on a least square method, performing second-order fitting on each pedestrian history track coordinate point to obtain fitting errors, initializing the fitting errors, and inputting the initialized fitting errors into a linear layer with random bias to obtain each pedestrian intention change probability vector;
inputting a random intention vector and an intention change probability vector into a sparse intention module, judging an element-by-element threshold value of the intention change probability by the sparse intention module, determining that a pedestrian uses the random intention vector or uses a zero vector without any influence to obtain an output vector, and completing construction of the random intention module;
and obtaining a displacement vector of the pedestrian by encoding the historical track coordinates of the pedestrian, and inputting a baseline model after superposing the output vector of the random intention module and the displacement vector of the pedestrian to obtain a pedestrian prediction track.
Further, the step of inputting the displacement information of the last frame into a linear layer with offset, and multiplying the obtained result by a random number factor to obtain a random intention vector includes the following steps:
obtaining the displacement of the last frame through the historical track coordinates, and then using one coordinate component of the displacement of the last frameAs reference information;
obtaining random number factors from standard normal distribution samplingWill->After feeding a linear layer with bias, result and +.>The random intention vector is obtained by multiplication:
wherein the method comprises the steps ofRepresenting multiplication of two vector corresponding elements, w and b representing learned weight parameters and bias terms in Linear layer Linear, ++>Contains the speed and direction of the pedestrian in the last frame, i.e. the intention information.
Further, the pedestrian track prediction method based on the intention randomness influence strategy provided by the invention further comprises the following steps:
after the random intention vector is obtained, combining a random number factorThe random intention vector combined with the current pedestrian state is preliminarily obtained, and then the values of w and b are continuously updated during model training, so that the model can realize self-regulation of the randomness in the random intention vector, and extremely unreasonable intention change is avoided.
Further, the step of inputting the initialized fitting error into a linear layer with random bias to obtain a probability vector for each pedestrian intention to change comprises the following steps:
performing second-order fitting on the historical track coordinate points of each pedestrian based on a least square method to obtain a fitting error;
initializing fitting errors into vectors with values between 0 and 1 according to the error magnitude
VectorEach represented by a Linear layer Linear with random biasThe vector of pedestrian intent change probabilities is:
where v and μ are learnable parameters in the random bias Linear layer Linear,is a set of random numbers extracted from the uniform distribution of interval [0, 1); weight v and random bias term +.>Aiming at different real scenes, the intention change probability of pedestrians is flexibly updated in the training process;
the linear layer uses an artificially designed random bias termThe pedestrian with small track fitting error in the network training process is considered to have intention change.
Further, the pedestrian track prediction method based on the intention randomness influence strategy provided by the invention further comprises the following steps:
after the pedestrian intention changing probability vector is obtained, training the pedestrian intention changing probability vector based on an intention loss function, so that flexible updating of the pedestrian intention changing probability is realized.
Further, the intent loss function is:
wherein,representing the true value of the intention change probability obtained according to the fitting error of the last frame of the historical track of the pedestrian and the future track;
the final loss function is defined as the baseline model loss function L Baseline And L Intention Linear combinations of (a), namely:
L＝L Baseline +λL Intention ；
wherein the penalty factor λ=0.1 is used to balance the primary-secondary relationship between the two losses.
Further, the sparse intention module performs element-by-element threshold judgment on the intention change probability, determines that a pedestrian uses a random intention vector or uses a zero vector without any influence, and obtains an output vector, and the method comprises the following steps:
using sparse intent modules, pairPerforming element-by-element threshold judgment, and generating a random number theta between 0 and 1 for each pedestrian n n ；
When (when)When (I)>The nth element of (2) is set to +.>Otherwise set to 0, i.e.:
finally generating a zero vector and a sparse random intention vectorSplicing to obtain module output->
Further, the encoding the historical track coordinates of the pedestrian to obtain a displacement vector of the pedestrian, and overlapping the output vector of the random intention module with the displacement vector of the pedestrian and inputting the overlapped result into the baseline model to obtain a predicted track of the pedestrian, which specifically comprises the following steps:
encoding historical track coordinates of pedestrians as displacement vectorsThe displacement is the coordinate difference between two adjacent frames, so that the input form of the baseline model data is ensured to be consistent with that of the tested baseline model data;
inputting the historical track coordinates of the pedestrians to a random intention module to obtain an intention module output spliced with the zero vector
For displacement vectorAnd intent module output->Performing superposition operation to obtain->Namely:
where ∈ represents element-by-element addition;
the influence of the change of the intention of the pedestrian is simulated by the change of the speed and the direction of the pedestrian caused by the intention vector.
Compared with the prior art, the pedestrian track prediction method based on the intention randomness influence strategy has the beneficial effects that:
the invention provides a random intention vector construction strategy, which combines history track information with random factors, and can directly consider the change of speed and direction caused by the intention change of pedestrians when predicting tracks; the invention designs the intention change probability by using the pedestrian history track fitting error, introduces the intention loss function self-updating intention change probability, and can pointedly model the change on the track caused by the pedestrian motion randomness; the method for constructing the intention vector can be used as a plug-in to be embedded into all track prediction reference methods, and supplements modeling of randomness of pedestrians so as to improve the accuracy of pedestrian track prediction of the reference methods.
Drawings
FIG. 1 is a schematic diagram of a random intention module provided by an embodiment of the present invention;
fig. 2 is a detailed diagram of implementation of a pedestrian track prediction baseline model provided by an embodiment of the invention.
Detailed Description
Embodiments of the present invention will be further described with reference to fig. 1 to 2. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
Example 1: as shown in fig. 1-2, the pedestrian track prediction method based on the intention randomness influence strategy provided by the invention specifically comprises the following steps:
1. construction of random intent vectors and intent change probabilities
The track of the first eight frames is taken as a historical track, so that the displacement of the eighth frames reflects the current speed direction and the current speed of the pedestrian, namely the current intention of the pedestrian is directly expressed, and the intention of the pedestrian is changed in a real scene to be visually expressed as the change on the track. And as soon as any velocity component of the pedestrian changes in the physical context, a subsequent change in velocity magnitude and direction will occur.
First the displacement of the 8 th frame obtained by the historical track coordinates, then use one of the coordinate components of the 8 th frame displacementAs reference information, will->Simulating the change amount of the pedestrian speed by combining some random number factors; in social context, the track of most pedestrians is basically unchanged or has small change amplitude, and few pedestrians have large change on the track; meanwhile, the simulation of the change of the track of the pedestrian needs to be capable of carrying out forward and reverse operations on the original state, so that the generated random number factor can preferably contain positive and negative number distribution; the present invention uses standard normal distribution to generate random number factor +.>
Obtaining random number factors from standard normal distribution samplingWill->After feeding a linear layer with bias, result and +.>The random intention vector is obtained by multiplication:
wherein the method comprises the steps ofRepresenting multiplication of two vector corresponding elements, w and b representing a learnable weight parameter and bias term in Linear layer Linear, ++>Comprising information of the speed and direction of the pedestrian in the last frame, i.e. intention, combined with a random number factor +.>The randomness brought can be preliminarily obtained by combining the current pedestrian stateThe random intention vector is updated continuously during model training, so that the model can realize self-adjustment of randomness in the random intention vector, and extremely unreasonable intention changes are avoided.
How to set a criterion from the historical track data to define which pedestrians are likely to be more prone to change in intent will make the random intent vector more targeted, so that the randomness of pedestrians is improved without too much influence. Because the track of the pedestrian is the visual reflection of the intention of the pedestrian, a fitting error module is designed, first, the second order fitting is carried out on the historical track coordinate points of each pedestrian based on the least square method to obtain the fitting error, and then the fitting error is initialized to be a vector with the value between 0 and 1 according to the error sizeThe vector is then +.>Through a Linear layer with random bias * The vector representing the probability of each pedestrian intent change is found as:
where v and μ are our defined random bias Linear layer Linear * Is used to determine the parameter of the model,is a set of random numbers extracted from the uniform distribution of intervals 0, 1). Here the weight v and the random bias term +.>Mainly for the purpose of updating the probability of the intention change of the pedestrian during training, while the linear layer here uses an artificially designed random bias term +.>Also, in order to enable pedestrians with small track fitting errors in the network training process to be considered to have intention change.
2. Intent loss function
Using only the loss function of the baseline model, the pedestrian intent-to-change probability vector cannot be effectively updatedTherefore, in order to guide the model to capture the relation between the historical track offset and the future track offset of the pedestrian, the invention introduces a new intention loss so as to ensure that the model can update the intention change probability +_ of the pedestrian in the training process more flexibly>The following is shown:
wherein the method comprises the steps ofRepresenting the true value of the intention change probability obtained by the fitting error of the last frame of the historical track of the pedestrian and the future track, and defining a final loss function as a loss function L of a baseline model Baseline And L Intention Linear combination of (i.e. l=l) Baseline +λL Intention We set a penalty factor λ=0.1 to balance the primary-secondary relationship between the two losses.
3. Sparse intent design
The random intention vector that has been obtainedAnd intent variation probability vector +.>Also by a sparse intention module to get sparse random intention vector +.>Sparse modules, i.e. according to +.>To decide whether its corresponding pedestrian uses the generated random intention vector or uses a zero vector without any influence, i.e. needs to be on +.>Making element-by-element threshold decisions, where no specific threshold is set, but a random number θ between 0 and 1 is generated for each pedestrian n n . Thus is->When (I)>The nth element of (2) is set to +.>Otherwise set to 0, i.e.:
finally, generating a zero vector and a sparse random intention vectorSplicing to obtain module output->So that the method is mainly used for conveniently superposing the displacement of the first eight frames which are complete with pedestrians, on one hand, the displacement of the first seven frames can not be influenced, and meanwhile, the method can also be used forTo update the weights in the intent vector at the time of training.
4. Baseline model implementation details
To evaluate the versatility of the present invention in constructing a random intent vector method, the above method is added as a module to existing methods for evaluation. As shown in fig. 2, first the historical track coordinates of the pedestrian need to be encoded as displacement vectorsThe displacement is defined as the difference in coordinates between two adjacent frames, which is for convenience consistent with the baseline model data input form of the test. Meanwhile, the historical track coordinates of the pedestrians are input into a random intention module to obtain the output of the intention module spliced with the zero vectorThen +.>And intent module output->Performing superposition operation to obtain->Namely:
where +.f. represents element-wise addition, this allows the intent vector we design to bring about changes in pedestrian speed and direction to simulate the effects of changes in pedestrian intent.
Example 2: in order to evaluate the effectiveness of the pedestrian trajectory prediction method provided in example 1 of the present invention, experiments were performed on two representative GNN and RNN-based advanced methods SGCN and STGAT official provided codes, as shown in table 1, the results of the baseline model and the addition of the present invention were implementedThe intelion-SGCN and intelion-STGAT following the pedestrian trajectory prediction method provided in example 1 were compared. In table 1, the results reproduced using the code provided by the authorities are not very identical to the results in the original text, possibly due to different settings of some super parameters or different experimental facilities. Therefore, for fair comparison and better reflecting the actual effect of the pedestrian trajectory prediction method provided in embodiment 1 of the present invention, all the super parameters and experimental equipment in evaluating the intellectualization-SGCN and intellectualization-STGAT are consistent with those in reproducing the baseline model, it can be seen that the Intention vector construction method of the present invention can improve the error index of the model test for different baseline models, wherein the average ADE/FDE and SGCN under 5 real scenes of the intellectualization-SGCN * Compared with the method for realizing the improvement of 0.02/0.12, the method for realizing the interaction-STGAT and the STGAT * An improvement of 0.03/0.05 is achieved; that is, by the evaluation results on several advanced baseline models, it can be seen that the pedestrian trajectory prediction method provided in embodiment 1 of the present invention is substantially improved in terms of actual error evaluation for both of these different types of methods. Where both ADE and FDE are better the lower.
TABLE 1
It can be seen that the pedestrian track prediction method based on the intention randomness influence strategy provided by the invention has the following beneficial effects:
(1) Experimental results on a plurality of large-scale public pedestrian track prediction data sets show that the performance of baseline pedestrian track prediction can be remarkably improved by the method;
(2) The proposed random intention vector construction strategy can guide the model to capture random intention changes of pedestrians, reduce burst influence caused by uncertainty of pedestrian movement, effectively improve the prediction efficiency of pedestrian trajectories, and have effectiveness in different real scenes;
(3) The setting of the intention change probability enables the random intention vectors to be more specific, so that the randomness of pedestrians is perfected, and meanwhile, the influence of excessive random intention vectors is not brought. A new intention loss is introduced on the basis of a baseline model loss function, so that the model can update the intention change probability of pedestrians in a training process more flexibly.
(4) The model provided by the invention can be added into a baseline model, training is carried out on a large-scale public pedestrian track prediction data set, and the trained weight is saved. The trained models and weights are then transplanted to the deep learning development board. When the automobile is loaded with the development board, the future track of surrounding pedestrians can be automatically judged by combining with environment perception, so that the driving route can be better planned in an auxiliary manner, and traffic accidents are reduced.
The above embodiments are merely preferred embodiments of the present invention, the protection scope of the present invention is not limited thereto, and any simple changes or equivalent substitutions of technical solutions that can be obviously obtained by those skilled in the art within the technical scope of the present invention disclosed in the present invention belong to the protection scope of the present invention.
Claims (7)
1. The pedestrian track prediction method based on the intention randomness influence strategy is characterized by comprising the following steps of:
extracting displacement information of the last frame from the historical track of the pedestrian, wherein the displacement information comprises the current speed direction and the current speed of the pedestrian;
obtaining a random number factor from standard normal distribution sampling for simulating the change amount of pedestrian speed;
inputting the displacement information of the last frame into a linear layer with offset, and multiplying the obtained result by a random number factor to obtain a random intention vector;
based on a least square method, performing second-order fitting on each pedestrian history track coordinate point to obtain fitting errors, initializing the fitting errors, and inputting the initialized fitting errors into a linear layer with random bias to obtain each pedestrian intention change probability vector;
inputting a random intention vector and an intention change probability vector into a sparse intention module, judging an element-by-element threshold value of the intention change probability by the sparse intention module, determining that a pedestrian uses the random intention vector or uses a zero vector without any influence to obtain an output vector, and completing construction of the random intention module;
the historical track coordinates of the pedestrians are coded to obtain displacement vectors of the pedestrians, and the output vectors of the random intention modules and the displacement vectors of the pedestrians are overlapped and then input into a baseline model to obtain predicted tracks of the pedestrians;
the displacement information of the last frame is input into a linear layer with offset, and the obtained result is multiplied by a random number factor to obtain a random intention vector, and the method comprises the following steps:
obtaining the displacement of the last frame through the historical track coordinates, and then using one coordinate component of the displacement of the last frameAs reference information;
obtaining random number factors from standard normal distribution samplingWill->After feeding a linear layer with bias, the result is summedThe random intention vector is obtained by multiplication:
wherein the method comprises the steps ofRepresenting multiplication of two vector corresponding elements, w and b representing the weights learned in Linear layer LinearHeavy parameters and bias terms->Contains the speed and direction of the pedestrian in the last frame, i.e. the intention information.
2. The pedestrian trajectory prediction method based on the intention randomness impact strategy as claimed in claim 1, further comprising:
after the random intention vector is obtained, combining a random number factorThe random intention vector combined with the current pedestrian state is preliminarily obtained, and then the values of w and b are continuously updated during model training, so that the model can realize self-regulation of the randomness in the random intention vector, and extremely unreasonable intention change is avoided.
3. The pedestrian trajectory prediction method based on the intention randomness impact strategy as claimed in claim 2, wherein the step of inputting the initialized fitting error into a linear layer with random bias to obtain each pedestrian intention change probability vector comprises the steps of:
performing second-order fitting on the historical track coordinate points of each pedestrian based on a least square method to obtain a fitting error;
initializing fitting errors into vectors with values between 0 and 1 according to the error magnitude
VectorThe vector representing the probability of each pedestrian intent change is obtained by passing a Linear layer Linear with random bias as follows:
where v and μ are learnable parameters in the random bias Linear layer Linear,is a set of random numbers extracted from the uniform distribution of interval [0, 1); weight v and random bias term +.>Aiming at different real scenes, the intention change probability of pedestrians is flexibly updated in the training process;
the linear layer uses an artificially designed random bias termThe pedestrian with small track fitting error in the network training process is considered to have intention change.
4. The pedestrian trajectory prediction method based on the intention randomness impact strategy as claimed in claim 3, further comprising:
after the pedestrian intention changing probability vector is obtained, training the pedestrian intention changing probability vector based on an intention loss function, so that flexible updating of the pedestrian intention changing probability is realized.
5. The pedestrian trajectory prediction method based on the intent randomness impact strategy as claimed in claim 4, wherein the intent loss function is:
wherein,representing the true value of the intention change probability obtained according to the fitting error of the last frame of the historical track of the pedestrian and the future track;
the final loss function is defined as the baseline model loss function L Baseline And L Intention Linear combinations of (a), namely:
L＝L Baseline +λL Intention ；
wherein the penalty factor λ=0.1 is used to balance the primary-secondary relationship between the two losses.
6. The pedestrian trajectory prediction method based on the intention randomness impact strategy as claimed in claim 5, wherein the sparse intention module performs element-by-element threshold judgment on the intention change probability, determines that the pedestrian uses a random intention vector or uses a zero vector without any influence, and obtains an output vector, and includes the following steps:
using sparse intent modules, pairPerforming element-by-element threshold judgment, and generating a random number theta between 0 and 1 for each pedestrian n n ；
When (when)When (I)>The element represented by the nth pedestrian of (2) is set to +.>The element represented by the nth pedestrian of (2), otherwise set to 0, i.e.:
finally generate a zero directionQuantity and sparse random intent vectorSplicing to obtain the output vector of the random intention module>
7. The pedestrian track prediction method based on the intention randomness impact strategy according to claim 6, wherein the step of obtaining the pedestrian predicted track by encoding the historical track coordinates of the pedestrian to obtain the displacement vector of the pedestrian, and adding the output vector of the random intention module and the displacement vector of the pedestrian to the baseline model, comprises the following specific steps:
encoding historical track coordinates of pedestrians as displacement vectorsThe displacement is the coordinate difference between two adjacent frames, so that the input form of the baseline model data is ensured to be consistent with that of the tested baseline model data;
inputting the historical track coordinates of the pedestrians to the random intention module to obtain an output vector of the random intention module spliced with the zero vector
For displacement vectorAnd the output vector of the random intention module +.>Performing superposition operation to obtain->Namely:
wherein the method comprises the steps ofRepresenting element-by-element additions;
the influence of the change of the intention of the pedestrian is simulated by the change of the speed and the direction of the pedestrian caused by the intention vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310130465.8A CN116259176B (en) | 2023-02-17 | 2023-02-17 | Pedestrian track prediction method based on intention randomness influence strategy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310130465.8A CN116259176B (en) | 2023-02-17 | 2023-02-17 | Pedestrian track prediction method based on intention randomness influence strategy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116259176A CN116259176A (en) | 2023-06-13 |
| CN116259176B true CN116259176B (en) | 2024-02-06 |
Family
ID=86685923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310130465.8A Active CN116259176B (en) | 2023-02-17 | 2023-02-17 | Pedestrian track prediction method based on intention randomness influence strategy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116259176B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929841A (en) * | 2019-10-11 | 2020-03-27 | 浙江大学城市学院 | Method for predicting pedestrian track based on circular neural network model with enhanced intention |
| CN111091581A (en) * | 2018-10-24 | 2020-05-01 | 百度在线网络技术（北京）有限公司 | Pedestrian trajectory simulation method and device based on generation of countermeasure network and storage medium |
| CN113538520A (en) * | 2021-08-02 | 2021-10-22 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and storage medium |
| CN113689470A (en) * | 2021-09-02 | 2021-11-23 | 重庆大学 | Pedestrian motion trajectory prediction method under multi-scene fusion |
| CN114626598A (en) * | 2022-03-08 | 2022-06-14 | 南京航空航天大学 | Multi-modal trajectory prediction method based on semantic environment modeling |
| CN114898550A (en) * | 2022-03-16 | 2022-08-12 | 清华大学 | Pedestrian trajectory prediction method and system |
| CN115272712A (en) * | 2022-07-26 | 2022-11-01 | 北京邮电大学 | Pedestrian trajectory prediction method fusing moving target analysis |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9786177B2 (en) * | 2015-04-10 | 2017-10-10 | Honda Motor Co., Ltd. | Pedestrian path predictions |
| EP4036892A1 (en) * | 2021-02-02 | 2022-08-03 | Aptiv Technologies Limited | Detection system for predicting information on pedestrian |
| KR102537381B1 (en) * | 2021-04-01 | 2023-05-30 | 광주과학기술원 | Pedestrian trajectory prediction apparatus |
-
2023
- 2023-02-17 CN CN202310130465.8A patent/CN116259176B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091581A (en) * | 2018-10-24 | 2020-05-01 | 百度在线网络技术（北京）有限公司 | Pedestrian trajectory simulation method and device based on generation of countermeasure network and storage medium |
| CN110929841A (en) * | 2019-10-11 | 2020-03-27 | 浙江大学城市学院 | Method for predicting pedestrian track based on circular neural network model with enhanced intention |
| CN113538520A (en) * | 2021-08-02 | 2021-10-22 | 北京易航远智科技有限公司 | Pedestrian trajectory prediction method and device, electronic equipment and storage medium |
| CN113689470A (en) * | 2021-09-02 | 2021-11-23 | 重庆大学 | Pedestrian motion trajectory prediction method under multi-scene fusion |
| CN114626598A (en) * | 2022-03-08 | 2022-06-14 | 南京航空航天大学 | Multi-modal trajectory prediction method based on semantic environment modeling |
| CN114898550A (en) * | 2022-03-16 | 2022-08-12 | 清华大学 | Pedestrian trajectory prediction method and system |
| CN115272712A (en) * | 2022-07-26 | 2022-11-01 | 北京邮电大学 | Pedestrian trajectory prediction method fusing moving target analysis |
Non-Patent Citations (1)
| Title |
|---|
| 基于社会力的自动驾驶汽车行人轨迹预测模型;杨文彦;张希;陈浩;金文强;;公路交通科技(08);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116259176A (en) | 2023-06-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112668153B (en) | Method, device and equipment for generating automatic driving simulation scene | |
| Al-Qizwini et al. | Deep learning algorithm for autonomous driving using googlenet | |
| CN111841012B (en) | Automatic driving simulation system and test resource library construction method thereof | |
| Feng et al. | Trafficgen: Learning to generate diverse and realistic traffic scenarios | |
| Essa et al. | Simulated traffic conflicts: do they accurately represent field-measured conflicts? | |
| CN107168303A (en) | A kind of automatic Pilot method and device of automobile | |
| CN112784485B (en) | Automatic driving key scene generation method based on reinforcement learning | |
| CN110716529A (en) | Automatic generation method and device for automatic driving test case | |
| CN110764494A (en) | Human-computer hybrid driving automatic driving virtual test platform based on SCANeR | |
| Alnaser et al. | Autonomous vehicles scenario testing framework and model of computation | |
| Thunyapoo et al. | Self-parking car simulation using reinforcement learning approach for moderate complexity parking scenario | |
| CN118228612B (en) | Natural automatic driving scene generation method and device based on reinforcement learning | |
| CN115291515A (en) | Automatic driving simulation test system and evaluation method based on digital twinning | |
| Gambi et al. | Automatically reconstructing car crashes from police reports for testing self-driving cars | |
| CN117521389A (en) | Vehicle perception test method based on vehicle-road collaborative perception simulation platform | |
| CN116685955A (en) | Method, device, electronic apparatus and medium for automatic driving system | |
| CN116259176B (en) | Pedestrian track prediction method based on intention randomness influence strategy | |
| Li | A scenario-based development framework for autonomous driving | |
| US20230281424A1 (en) | Method for Extracting Features from Data of Traffic Scenario Based on Graph Neural Network | |
| Tosin et al. | Control strategies for road risk mitigation in kinetic traffic modelling | |
| Tian et al. | An LLM-enhanced Multi-objective Evolutionary Search for Autonomous Driving Test Scenario Generation | |
| Zhang et al. | A method for evaluating the complexity of test scenarios for autonomous vehicles | |
| Desta et al. | Simulation experiments and validation for considering bus priority treatments in small cites: The case of Hawassa city, Ethiopia | |
| Pariota et al. | Motivating the need for an integrated software architecture for connected and automated vehicles technologies development and testing | |
| Thorat et al. | Modelling and Simulation-Based Design Approach to Assisted and Automated Driving Systems Development |
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 |