Fused automatic driving automobile street-crossing pedestrian trajectory prediction method and system

Abstract

The invention provides a fused automatic driving automobile street-crossing pedestrian trajectory prediction method and system, and the method comprises the steps: obtaining street-crossing pedestrianmotion state information, pedestrian individual feature information and vehicle motion state information according to a vehicle-mounted sensor fusion algorithm; calibrating parameters in the social force model according to the state data obtained by the vehicle-mounted sensor; training structural weight and offset parameters of an LSTM model according to the state data obtained by the vehicle-mounted sensor; utilizing the social force model and the LSTM model to respectively predict the motion trails of pedestrians crossing the street; importing the motion trail predicted by the model and theactual trail truth value of the pedestrian crossing the street into a Stacking fusion model, and training a structure weight; and outputting the optimal prediction track of the pedestrian crossing the street within a first preset time length in the future by utilizing the Stacking fusion model. According to the method, the Stacking algorithm is used for fusing the social force model and the LSTMmodel, the effect of reducing variance and deviation is achieved, and therefore it is guaranteed that the predicted trajectory is closer to the actual trajectory of the pedestrian.

Description

technical field [0001] The present invention relates to the field of decision-making algorithms for automatic driving, in particular, to a method and system for predicting trajectories of pedestrians crossing the street for autonomous vehicles, and more specifically, to a trajectory of pedestrians crossing streets for autonomous vehicles, which integrates the social force model and the LSTM model The prediction method involves a whole set of processes from the preparatory work to the specific implementation method in the later stage. Background technique [0002] In recent years, with the rapid development of autonomous vehicle technology, pedestrian safety protection is an important factor that must be considered by autonomous vehicles. Pedestrians, as the main participants in traffic, are more complex and changeable than vehicle movements, and have great flexibility and randomness. For autonomous vehicles, understanding pedestrian behavior and predicting trajectories is a ...

AI Technical Summary

Problems solved by technology

The overly conservative intelligent decision-making behavior of self-driving cars in the face of pedestrians not only reduces the traffic capacity of the road, but may even cause road congestion

[0004] Existing pedestrian trajectory prediction methods, such as Momenta, currently use the simplest constant velocity (CV) model or constant acceleration (CA) model to predict the trajectory of pedestrians, in order to remind the driver whether there will be pedestrians suddenly breaking into the driving area. Accidents, but the model has low accuracy and poor implementation effect, so it is not suitable for self-driving cars that require extremely high accuracy

There is also a dynamic Bayesian network (DBN) used to predict the movement state of pedestrians crossing the street when facing an approaching vehicle (stop and go prediction). Efficiency; there is also the use of deep learning long-short-term memory network (LSTM) model to predict the trajectory of pedestrians. Although the prediction effect is good, it only considers the trajectory prediction of a single pedestrian in isolation, and does not consider vehicles, surrounding pedestrians and other traffic environments. Impact on Target Pedestrians

[0005] Although many achievements have been made in the trajectory prediction of pedestrians, the main problem in the prediction process of existing methods is that pedestrians are predicted as general obstacles, and the surrounding traffic environment cannot be considered from a sociological point of view. The impact on target pedestrians, such as the impact of surrounding pedestrians on target pedestrians, the impact of vehicles on target pedestrians, and the impact of traffic lights and zebra crossings on pedestrians, etc., will result in poor accuracy of predicted pedestrian trajectories

In addition, the influence of individual differences of pedestrians is rarely considered

At present, there is no pedestrian trajectory prediction method that considers factors such as the surrounding traffic environment and pedestrian personality differences.

Images