Intelligent vehicle risk assessment method based on vehicle trajectory prediction

Abstract

The invention discloses an intelligent vehicle risk assessment method based on vehicle trajectory prediction. The method comprises a trajectory prediction step and a threat assessment step. In the trajectory prediction step, a deep learning algorithm is adopted, an NGSIM data set training and evaluation behavior prediction model is utilized to realize driving intention classification, and trajectory probability distribution of the autonomous vehicle is obtained; and in the threat assessment step, the threat in each state is quantitatively and objectively assessed through a vehicle risk assessment function, and the vehicle risk assessment function comprises collision time TTC, vehicle head time interval TH and enhanced collision time ETTC. Trajectory prediction and risk assessment of the vehicle are combined for consideration, a complete model structure is established, it is guaranteed that after the trajectories of the autonomous vehicle and the surrounding vehicles are obtained, whether the trajectories conflict with the traffic vehicle or not can be assessed, and then whether the autonomous vehicle is dangerous or not is determined.

Description

technical field [0001] The invention relates to the technical field of automatic driving, in particular to an intelligent vehicle risk assessment method based on its vehicle trajectory prediction. Background technique [0002] With the improvement of the level of autonomous driving, people are fully aware that the decision-making and trajectory planning of smart cars need to be based on a full understanding of the surrounding traffic environment, that is, the vehicle needs to constantly perceive and understand the surrounding environment, and then Only by making certain predictions about the development of future traffic situation can high-quality driving decisions and planning be made. This point has been paid attention to by many researchers, and a lot of research work has been invested and certain research results have been obtained, which focus on the following two aspects: 1. For the multi-lane environment, predict the driving trajectory of the vehicle around the main v...

AI Technical Summary

Problems solved by technology

[0006] The application number is CN202010248715.4 Chinese patent document, which is a safety assessment method based on deep convolutional neural network and intersection. The trajectory extraction of this method is mainly based on manual calibration, which requires a lot of manpower and material resources to acquire the trajectory, and Unable to implement to reflect risk level

[0007] The application number is CN202010088791.3 Chinese patent document, which proposes a collision risk assessment based on vehicle acceleration, which uses the trajectory extracted from the video to directly perform risk assessment, but in fact the complete risk assessment process includes trajectory prediction and risk assessment Two parts, the risk assessment without considering the trajectory prediction will lead to only focusing on the risk at a certain moment, but not taking into account the risk in a period of time

[0008] The application number is CN 201911233101.2 Chinese patent document, which is a driving risk assessment method. This method relies on communication between vehicles for risk assessment of unmanned vehicles, and only considers historical trajectories, not future trajectories of other vehicles. So this method will make the model less accurate and less robust

Images