Study On Multi-task And Multi-model Probabilistic Trajectory Prediction Method In Traffic Scenes

With the development of artificial intelligence technology,autonomous driving technology has become a research hotspot in the automotive industry,which is of great significance for improving traffic safety,increasing social and economic benefits.In the complex road traffic scene of multi-agent interaction,making reasonable decisions is an important prerequisite to ensure the safety and effectiveness of autonomous driving technology,and the prediction of multi-agent interaction trajectory is the key to determining the rationality of smart car decisionmaking.Based on the modeling ideas of State-Anchor and Anchor-Free,this paper carries out modeling and analysis of vehicles,cyclists and pedestrians,designing multi-task networks,and proposing multi-task and multi-model probability trajectory prediction model,it will lay the foundation for the subsequent behavioral decision-making and control of autonomous vehicles.First of all,in view of the problem that the data category of the current prediction data set is too single and the data samples are too few,this paper improves the trajectory prediction data set that contains the road high-definition map and the obstacle history sequence,and combines the environmental information and various agents in the scene.Rasterized representation provides rich semantic topology and historical sequence information for subsequent model modeling.After that,we use trajectories of vehicles,cyclists and pedestrians to cluster analysis,it will model the intent uncertainty and control uncertainty of vehicles,cyclists and pedestrians,and based on the modeling ideas of State-Anchor and Anchor-Free,we build a multi-model probability trajectory prediction model for vehicles,cyclists and pedestrians.Through the use of Encoder-Interaction-Decoder network architecture,we design the network model of vehicles,cyclists and pedestrians,design the model loss function and parameter selection.Finally,in order to improve the efficiency of model prediction,a multi-task multi-model probabilistic trajectory prediction model is proposed to realize the real-time parallel prediction of vehicles,cyclists and pedestrians in the scene.The model in this paper is compared with the current mainstream single-category models quantitatively and qualitatively to demonstrate the superiority of our model.Visual analysis of model prediction results shows that the multi-task model can predict multiple reasonable trajectory sequences for each agent,and provide multiple probabilistic alternative trajectories for subsequent smart cars to make more anthropomorphic decisions.