Research And Implementation Of Trajectory Prediction Technology For Complex Scenes

With the development of science and technology,a variety of positioning devices continue to emerge,and the number and types of trajectory data that can be obtained are rapidly increasing.The storage and analysis of the acquired trajectory data can play an important role in behavior recognition,traffic planning,urban safety and prevention and control.The study of human trajectories can obtain key information such as behavior patterns and personal preferences,which can promote further research in many fields.Therefore,it is of great significance to study the pedestrian trajectory prediction algorithm.Traditional trajectory prediction methods are based on mathematical statistical models to model pedestrian movement patterns,which are difficult to apply to complex scenes.In complex scenes,there are interference factors such as obstacles,large numbers of pedestrians and changes in relationships.The existing methods are difficult to effectively deal with the influence of these factors on the prediction process.At the same time,because trajectory prediction tasks often need to predict multiple moments,the accumulation of errors in the prediction process also requires effective solutions.To this end,this thesis proposes a new trajectory prediction framework that can learn the changes in the relationship between pedestrians during the prediction process,and at the same time reduce the impact of error accumulation on the prediction results during the prediction process.The main contributions of this thesis are as follows:(1)This thesis proposes a trajectory prediction method based on the attention mechanism.Aiming at the problem that it is difficult to effectively use the long-interval time sequence features in the trajectory prediction task,this thesis proposes a trajectory prediction method based on the attention mechanism,which highlights the trajectory characteristics of key moments at the feature level,and calculates the influence coefficients of different historical time points on the predicted time points.The trajectory characteristics of historical moments are integrated to reduce the influence of error accumulation phenomenon on the prediction results during the prediction process.(2)This thesis proposes a trajectory prediction method based on adaptive graph convolutional network.In response to the complex and changeable relationship among pedestrians in the trajectory prediction task,it is difficult to accurately model pedestrian interaction.On the basis of previous work,a pedestrian interaction modeling method based on adaptive graph convolution network is added,and adaptive graph convolution is adopted.The network models pedestrian interaction,study the correlation between pedestrians,integrate the characteristics of pedestrian trajectory,and carry out targeted experiments to prove this method.(3)This thesis implements a pedestrian trajectory prediction system.In order to verify the effectiveness of the method in this thesis,this thesis designs and implements a pedestrian trajectory prediction system based on actual needs to meet the functional needs of users and provide users with video pedestrian trajectory prediction functions.