Research On Multi-sensor Multi-target Tracking And Information Fusion Of Intelligent Vehicles

Autonomous driving technology had great potential to reduce traffic accidents and alleviated traffic congestion.It has become a research hotspot in the current automotive industry.Among the many research directions of autonomous driving,stable tracking of the targets by the vehicle sensor network(VSN),and accurate information fusion of targets obtained by the VSN were the prerequisites for car to make correct decisions,and these were also important guarantees for safe driving.There were two difficulties in the process of multi-sensor information fusion by the VSN.One was how to accurately estimate the state of targets and perform stable tracking of targets.Another was how to accurately and effectively correlate and fuse the target from different sensors to get a comprehensive description of the target.This paper focused on the information fusion methods of multiple heterogeneous sensors,mainly including data association and non-linear state estimation methods,as well as target correlation and fusion methods in the VSN.The effect of these methods was verified by virtual environment simulation.The main research contents of this paper are as follows:1.In order to deal with the problem of multi-target tracking data association,this paper introduced the basic principles of multi-target tracking and derived the target motion models.Then it compared the Joint Probabilistic Data Association with the Multiple Hypothesis Tracking.2.In order to deal with the problem of multi-target tracking state estimation problem,this paper introduced the Bayesian filtering principle,which considered that the process of state estimation is non-linear.In practical situations,the nonlinear optimal filtering was difficult to implement.So,three types of suboptimal filtering methods were introduced,i.e.extended Kalman filtering,unscented Kalman filtering and cubature Kalman filtering,particle filtering.To improve the numerical accuracy and stability of nonlinear state estimation algorithms in engineering applications,this paper proposed a square root cubature joint probability data association multi-target tracking algorithm.Firstly,in order to overcome the filter divergence caused by rounding errors,reduce the computational complexity,ensure the non-negative definiteness of covariance,and improve the convergence speed of the filter,the joint probability data association equipped with the square root version cubature Kalman filter was deployed;Secondly,in order to reduce the computational complexity of the data association algorithm,an adaptive tracking gate combined with the elliptical gate and vehicle kinematics of the vehicle is constructed.3.In order to deal with the problem of multi-sensor target correlation and fusion,this paper analyzes common track-to-track correlation and information fusion algorithms.A distributed correlation and fusion algorithm based on covariance is proposed.Independent sequential target correlation algorithm and weighted fusion algorithm were implemented to fuse the different sensor targets after coordinate transformation,which effectively improved the reliability and accuracy of the system.4.According to the method proposed in this paper,the camera and millimeter-wave radar fusion architecture was established,and simulation verification was performed in a virtual environment.The simulation traffic scene,sensor model,and vehicle model were built in Pre Scan,and the algorithm was implemented in Simulink.The algorithm was evaluated and analyzed with indicators such as root mean square error.The result shown that it was closer to the ground truth of the target by proposed method in this paper,which verifies the effectiveness of the algorithm.