Urban road vehicle trajectory prediction algorithm based on external knowledge enhancement
The integration of pedestrian trajectory data into vehicle prediction models through a distance memory network enhances prediction accuracy and reliability in urban environments by dynamically adjusting for pedestrian impacts, addressing the limitations of traditional algorithms.
Patent Information
- Application Number
- CN202411082624.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing vehicle trajectory prediction algorithms in urban environments struggle to accurately account for dynamic factors such as pedestrians and other dynamic factors, which existing algorithms mostly rely on the vehicle itself, ignoring the complex and changeable urban road environment, resulting in insufficient prediction accuracy.
A vehicle trajectory prediction model that integrates pedestrian trajectory information to improve prediction accuracy by incorporating pedestrian trajectory data through a distance memory network with a distance attention mechanism and correction mechanism, adjusting vehicle trajectories dynamically.
Enhances prediction accuracy and reliability by dynamically adjusting for pedestrian impacts, effectively modeling complex urban traffic interactions, and improving model stability.
Smart Images
Figure CN119152671A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of vehicle trajectory prediction, and in particular relates to an external knowledge enhanced urban road vehicle trajectory prediction algorithm. Background Art
[0002] With the acceleration of urbanization, the complexity of urban transportation systems is increasing day by day. Cellular Vehicle-to-Everything (C-V2X) technology has emerged in this context, aiming to improve traffic safety and efficiency. Among them, autonomous driving technology, as an important part of intelligent transportation, has received widespread attention. Vehicle trajectory prediction is a key link in realizing autonomous driving. By accurately predicting the future trajectory of the vehicle, it can effectively avoid collisions, optimize traffic flow, and improve driving safety.
[0003] Existing trajectory prediction algorithms are mostly based on the kinematic characteristics of the vehicle itself and use historical trajectory data for prediction. However, the urban road environment is complex and changeable, and factors such as pedestrians, traffic lights, and emergencies have a significant impact on vehicle trajectories. It is difficult to fully capture these dynamic changes by simply relying on the vehicle's own data. In practical applications, the urban road environment is complex and changeable, and pedestrians and other dynamic factors have a greater impact on vehicle trajectories. Traditional trajectory prediction algorithms mainly rely on the vehicle's own historical data and ignore the dynamic factors of pedestrians, resulting in insufficient prediction accuracy. Summary of the invention
[0004] Aiming at the technical problems existing in the prediction of vehicle trajectories on urban roads, the present invention proposes a vehicle trajectory prediction algorithm integrating pedestrian trajectory information, corrects and optimizes the vehicle trajectory, and designs a pedestrian trajectory penalty weight taking into account the activity patterns and behavior patterns of pedestrians on urban roads, so as to adjust the vehicle trajectory prediction model and make it more in line with the actual traffic scenes. An external knowledge-enhanced urban road vehicle trajectory prediction algorithm is proposed.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is: an external knowledge enhanced urban road vehicle trajectory prediction algorithm, comprising the following steps:
[0006] S1. Collect historical trajectory data of vehicles and pedestrians. The data format is two-dimensional coordinates in the form of time series, respectively (x v,t ,y v,t )、(x p,t ,y p,t );
[0007] S2. Feature extraction of vehicle trajectory data based on historical trajectory data of vehicles and pedestrians, including dynamic features of position, speed, and acceleration. The speed and acceleration of the vehicle are:
[0008]
[0009]
[0010] Among them, v t represents the speed of the vehicle at time t, a t represents the acceleration of the vehicle at time t, Δt represents the time interval, and the pedestrian trajectory data is feature extracted to extract the position and speed of the pedestrian:
[0011] p t =(x p,t ,y p,t )
[0012]
[0013] Among them, p t represents the position of the pedestrian at time t, v p,t represents the speed of the pedestrian at time t;
[0014] S3. Design a distance memory network model, which includes two parts: distance attention mechanism and distance correction mechanism. First, the distance attention mechanism calculation formula introduces the pedestrian trajectory penalty weight to obtain the pedestrian distance attention state. The distance attention mechanism vehicle trajectory calculation method is as follows:
[0015] f t (V t ,P t ,W)=σ(W f *[h t-1 ,V t ]+b f -W*P t )
[0016] Among them, W f and b f is the weight and bias of the distance attention mechanism, σ is the activation function, h t-1 is the hidden state information of the previous moment, V t ×(x v,t ,y v,t ,v t ,a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t. The distance correction mechanism does not refer to the prediction of the pedestrian trajectory to make a certain distance correction. The coordinate correction error value is calculated as follows:
[0017] o t (V t ,di,j ,W)=σ(W o *[h t-1 ,V t ]+b o )*(1-W*tanh(d i,j ))
[0018] Among them, W o and b o is the weight and bias of the output gate, h t-1 is the hidden state information of the previous moment, V t =(x v,t ,y v,t ,v t ,a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t. In addition, the activation function σ is calculated as follows:
[0019]
[0020] W represents the penalty weight of the pedestrian trajectory, and W has the following calculation formula:
[0021] W = α*exp(-β*d i,j )
[0022] Among them, α and β are hyperparameters, d i,j =||(x v,t -x p,t ,y v,t -y p,t )|| is the distance between the vehicle and the pedestrian, and the loss function of the distance memory network model is defined as follows:
[0023]
[0024] Where N represents the length of time, represents the predicted value of the vehicle trajectory at time t, y t represents the true value of the vehicle trajectory at time t, λ is a hyperparameter whose value range is (0,1), W t represents the trainable weights of the distance memory network model, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t;
[0025] S4. Correct the vehicle trajectory. The vehicle trajectory correction formula is as follows:
[0026]
[0027] in, represents the predicted position of the vehicle at time t+1, f t (V t ,P t ,W) is the distance attention mechanism vehicle trajectory, o t (V t ,d i,j ,W) is the coordinate correction error value;
[0028] Preferably, the activation function described in step S3 has an output close to the input value when the input value is small, and an output tends to 1 or -1 when the input value is large.
[0029] Compared with the prior art, the advantages and positive effects of the present invention are:
[0030] The urban road vehicle trajectory prediction algorithm based on the external knowledge enhancement of pedestrian trajectory information proposed in the present invention significantly improves the accuracy and reliability of trajectory prediction by integrating pedestrian trajectory information. In practical applications, the urban road environment is complex and changeable, and pedestrians and other dynamic factors have a great influence on vehicle trajectories. Traditional trajectory prediction algorithms mainly rely on the historical data of the vehicle itself, ignoring these dynamic factors, resulting in insufficient prediction accuracy. By introducing the penalty weight of pedestrian trajectories, this algorithm can dynamically adjust the impact of pedestrians on vehicle trajectories, reflecting the effective modeling of complex interactions in real traffic scenes. The new activation function design makes the model more stable when processing large inputs, avoids numerical problems caused by excessive output, and enhances the robustness of the model. The design of this method relies on high-quality data collection, so as the automatic driving data continues to be enriched, the model effect will become better and better. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0032] Figure 1 A schematic diagram of the structure of a distance memory network proposed in an embodiment of the present invention. DETAILED DESCRIPTION
[0033] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.
[0034] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.
[0035] Embodiment, considering that the existing trajectory prediction algorithms are mostly based on the kinematic characteristics of the vehicle itself, they use historical trajectory data for prediction. However, the urban road environment is complex and changeable, and factors such as pedestrians, traffic lights, and emergencies have a significant impact on the vehicle's trajectory. It is difficult to fully capture these dynamic changes by simply relying on the vehicle's own data. In actual applications, the urban road environment is complex and changeable, and pedestrians and other dynamic factors have a greater impact on the vehicle's trajectory. The traditional trajectory prediction algorithm mainly relies on the vehicle's own historical data and ignores the dynamic factors of pedestrians, resulting in insufficient prediction accuracy. For this reason, Figure 1 As shown in the figure, the present invention proposes an external knowledge enhanced urban road vehicle trajectory prediction algorithm. First, the historical trajectory data of vehicles and pedestrians are collected. The data format is a two-dimensional coordinate in the form of a time series, which are (x v,t ,y v,t )、(x p,t ,y p,t ).
[0036] Subsequently, the vehicle trajectory data is feature extracted based on the historical trajectory data of vehicles and pedestrians, including the dynamic characteristics of position, speed, and acceleration. The speed and acceleration of the vehicle are:
[0037]
[0038]
[0039] Among them, v t represents the speed of the vehicle at time t, a t represents the acceleration of the vehicle at time t, Δt represents the time interval, and the pedestrian trajectory data is feature extracted to extract the position and speed of the pedestrian:
[0040] p t =(x p,t ,y p,t )
[0041]
[0042] Among them, p t represents the position of the pedestrian at time t, v p,t represents the speed of the pedestrian at time t.
[0043] By introducing the penalty weight of pedestrian trajectories, the impact of pedestrians on vehicle trajectories can be dynamically adjusted, which reflects the effective modeling of complex interactions in real traffic scenes. So next, a distance memory network model is designed, which includes two parts: distance attention mechanism and distance correction mechanism. First, the distance attention mechanism calculation formula introduces the pedestrian trajectory penalty weight to obtain the pedestrian distance attention state. The distance attention mechanism vehicle trajectory calculation method is as follows:
[0044] f t (V t ,P t ,W)=σ(W f *[h t-1 ,V t ]+b f -W*P t )
[0045] Among them, W f and b f is the weight and bias of the distance attention mechanism, σ is the activation function, h t-1 is the hidden state information of the previous moment, V t =(x v,t ,y v,t ,v t ,a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t. The distance correction mechanism does not refer to the prediction of the pedestrian trajectory to make a certain distance correction. The coordinate correction error value is calculated as follows:
[0046] o t (V t ,d i,j ,W)=σ(W o *[h t-1 ,V t ]+b o )*(1-W*tanh(d i,j ))
[0047] Among them, W o and b o is the weight and bias of the output gate, h t-1 is the hidden state information of the previous moment, V t =(x v,t ,y v,t ,v t ,a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p,t ,y p,t,v p,t ) represents the trajectory characteristics of the pedestrian at time t. In addition, the activation function σ is calculated as follows:
[0048]
[0049] When the input value is small, the output of the activation function is close to the input value, and when the input value is large, the output tends to 1 or -1. This design can avoid excessive activation values while maintaining the input value amplitude, thereby enhancing the stability of the model. W represents the penalty weight of the pedestrian trajectory, and W has the following calculation formula:
[0050] W = α*exp(-β*d i,j )
[0051] Among them, α and β are hyperparameters, d i,j =||(x v,t -x p,t ,y v,t -y p,t )|| is the distance between the vehicle and the pedestrian, and the loss function of the distance memory network model is defined as follows:
[0052]
[0053] Where N represents the length of time, represents the predicted value of the vehicle trajectory at time t, y t represents the true value of the vehicle trajectory at time t, λ is a hyperparameter whose value range is (0,1), W t represents the trainable weights of the distance memory network model, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t. This algorithm can dynamically adjust the impact of pedestrians on vehicle trajectories, reflecting the effective modeling of complex interactions in real traffic scenarios.
[0054] Finally, the vehicle trajectory is corrected. The vehicle trajectory correction formula is as follows:
[0055]
[0056] in, represents the predicted position of the vehicle at time t+1, f t (V t ,P t ,W) is the distance attention mechanism vehicle trajectory, o t (V t ,d i,j ,W) is the coordinate correction error value;
[0057] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. An external knowledge enhanced urban road vehicle trajectory prediction algorithm, characterized in that: The steps include: S1. Collect historical trajectory data of vehicles and pedestrians. The data format is two-dimensional coordinates in the form of time series, respectively (x v,t ,y v,t )、(x p,t ,y p,t ); S2. Feature extraction of vehicle trajectory data based on historical trajectory data of vehicles and pedestrians, including dynamic features of position, speed, and acceleration. The speed and acceleration of the vehicle are: Among them, v t represents the speed of the vehicle at time t, a t represents the acceleration of the vehicle at time t, Δt represents the time interval, and the pedestrian trajectory data is feature extracted to extract the position and speed of the pedestrian: p t =(x p，t ,y p，t ) Among them, p t represents the position of the pedestrian at time t, v p，t represents the speed of the pedestrian at time t; S3. Design a distance memory network model, which includes two parts: distance attention mechanism and distance correction mechanism. First, the distance attention mechanism calculation formula introduces the pedestrian trajectory penalty weight to obtain the pedestrian distance attention state. The distance attention mechanism vehicle trajectory calculation method is as follows: f t (V t ,P t ,W)=σ(W f *[h t-1 ,V t ]+b f -W*P t ) Among them, W f and b f is the weight and bias of the distance attention mechanism, σ is the activation function, h t-1 is the hidden state information of the previous moment, V t =(x v，t ,y v，t , v t , a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p，t ,y p，t , v p，t ) represents the trajectory characteristics of the pedestrian at time t. The distance correction mechanism does not refer to the prediction of the pedestrian trajectory to make a certain distance correction. The coordinate correction error value is calculated as follows: o t (V t ,d i，j ,W)=σ(W o *[h t-1 ,V t ]+b o )*(1-W*tanh(d i，j )) Among them, W o and b o is the weight and bias of the output gate, h t-1 is the hidden state information of the previous moment, V t =(x v,t ,y v,t ,v t ,a t ) represents the trajectory characteristics of the vehicle at time t, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t. In addition, the activation function σ is calculated as follows: W represents the penalty weight of the pedestrian trajectory, and W has the following calculation formula: W=α*exp(-β*d i，j ) Among them, α and β are hyperparameters, d i,j =||(x v,t -x p,t ,y v,t -y p,t )|| is the distance between the vehicle and the pedestrian, and the loss function of the distance memory network model is defined as follows: Where N represents the length of time, represents the predicted value of the vehicle trajectory at time t, y t represents the true value of the vehicle trajectory at time t, λ is a hyperparameter whose value range is (0,1), W t represents the trainable weights of the distance memory network model, P t =(x p,t ,y p,t ,v p,t ) represents the trajectory characteristics of the pedestrian at time t; S4. Correct the vehicle trajectory. The vehicle trajectory correction formula is as follows: in, represents the predicted position of the vehicle at time t+1, f t (V t ,P t ,W) is the distance attention mechanism vehicle trajectory, o t (V t ,d i,j ,W) is the coordinate correction error value.
2. The external knowledge enhanced urban road vehicle trajectory prediction algorithm according to claim 1 is characterized in that: The activation function described in step S3 outputs a value close to the input value when the input value is small, and outputs a value close to 1 or -1 when the input value is large.