Random Finite Set Based Multi-target Tracking Method For Intelligent Vehicle

Target tracking is one of the key technologies of the intelligent vehicle environment perception system.Improving the accuracy and reliability of target tracking,environment recognition and decision-making control was effectively improved,for realizing automatic driving have an important role in complex traffic scenarios.Random Finite Set(RFS)uses a unified target tracking theoretical framework to improve target tracking performance in complex environments.It is one of the new ways to solve the problem of intelligent vehicle target tracking.Based on the RFS theory,the target tracking problem of intelligent vehicles in complex traffic scenarios.The main work content includes:1.The angle Association fusion tracking algorithm of radar and monocular camera in complex road environment is studies.In the intelligent vehicle target tracking system composed of monocular camera sensor and millimeter wave radar sensor,the ranging error of monocular camera is large and changes violently,while the radar echo is large and the false alarm is high.A fusion tracking algorithm based on angle value to correlate the measurement data of monocular camera and radar in polar coordinate system is proposed.The simulation results verify the effectiveness of the fusion algorithm.2.New target intensity estimation method of Gaussian mixture probability hypothesis density filtering algorithm is studied.In traditional GMPHD filtering algorithm,the strength of new targets is regarded as a priori information,but the position of new targets can not be determined in the process of tracking.An adaptive GM-PHD filter of new target strength is proposed.By classifying Gaussian components,the new targets and survival targets are distinguished,and the prediction and update steps of the two are separated.The new target intensity is calculated adaptively,which avoids the condition of a priori hypothesis,effectively reduces the amount of calculation and the tracking accuracy was improved.Simulation results show the effectiveness of the improved algorithm.3.The target tracking algorithm based on interacting multiple model and Ph D filter is studied.Usually,the target motion model is established as an independent motion model in the research of tracking algorithm,which does not consider the influence of the surrounding environment or other targets on the tracking target,so that the trajectory of maneuvering target can not be well predicted.In thesis,an interactive multiple model Gaussian mixture probability hypothesis density filtering algorithm for maneuvering target is proposed.The force of the front target on the rear target is introduced into the motion model as an additional acceleration term,and the independent motion model and interactive motion model are effectively switched through the interactive multiple model algorithm,so as to improve the tracking performance when the target maneuvers.The simulation results verify the effectiveness of the proposed method in estimation accuracy.