| Relying on advanced sensor perception,decision-making and execution technologies,Intelligent Connected Vehicle supports a new automobile civilization,which can effectively ensure traffic safety,reduce transportation costs,improve vehicle efficiency and reduce air contamination,and is an important development direction in the field of future Intelligent Transportation System.Among them,the perception technology of Intelligent Connected Vehicle for surrounding targets is a very important cornerstone of decision-making and execution technology.High-precision and blind-spot-free perception capabilities can further improve the safety level of vehicle.However,the traditional single vehicle multiple target tracking is easily affected by various factors such as obstacle occlusion,sensor blind spot,noise interference,etc.,and has problems such as low tracking accuracy and large driving blind spot.With the rapid development of the V2 X communication technology,multiple Intelligent Connected Vehicle can share local sensing information through wireless communication,and the sensing data of onboard sensors can be fused with data shared by other intelligent vehicles to achieve cooperative sensing,which is an effective way to improve the performance of multiple target tracking.Considering the uncertainty of sensor measurement noise and target vehicle motion,this paper relies on the national key research and development plan sub-project "Multiple target tracking method under multiple vehicle cooperative perception(2018YFB0105004)",under the framework of Bayesian filtering,from the perspectives of the uncertainty of the measurement model caused by the positioning data being easily interfered by outliers,and the dynamic changes of the motion model caused by the target mobility,the multiple target cooperative tracking method has been systematically studied,and the main work is summarized as follows:(1)Starting from the relevant theories and methods of multiple target tracking,the multiple target tracking based on data association and the multiple target tracking based on Random Finite Set are discussed in detail.The former includes multiple hypothesis tracking,multiple target tracking based on joint probability data association and multiple target tracking based on Hungarian algorithm;the latter includes Random Finite Set theory,multiple target Bayesian recursion and approximate multiple target Bayesian filter.Lay the foundation for the follow-up research.(2)The precise positioning of Intelligent Connected Vehicle can significantly improve the performance of multiple target cooperative tracking.However,due to factors such as multi-path effect,atmospheric disturbances,and obstructions,the positioning information of satellite navigation system leads to unknown statistical characteristics of positioning measurement noise and is easily disturbed by outliers,which in turn affects the tracking performance.Therefore,a Bayesian model for jointly estimating the target state and positioning noise parameters is established for the uncertainty of the measurement model in the multiple target cooperative tracking scenario.Aiming at the mutual coupling between target state and noise distribution parameters,a recursive estimation algorithm of system parameters based on variational Bayesian inference principle is proposed,and the variational posterior distribution is decoupled by means of mean field theory,and the effective calculation of the variational posterior distribution is robustly implemented in an alternate optimization manner.(3)An accurate target motion model can provide reliable prior information for state estimation.Aiming at the factors such as missed detection and false detection of measurement data and dynamic changes of the motion model caused by target mobility in complex scenarios,an adaptive multiple model probability hypothesis density filtering(AMMGMPHD)method based on Random Finite Set is proposed to estimate the global state of maneuvering multiple targets.Then,on the basis of estimating the global state of the multiple targets of the multiple Intelligent Connected Vehicle,a maneuvering multiple target cooperative tracking algorithm based on AMMGMPHD filtering is designed,and the Hungarian algorithm and the fast covariance intersection method are used to complete the association and fusion of the multiple target state estimation under the fusion central coordinate system,and obtain the maneuvering multiple target state estimation after cooperative tracking.(4)Complete pseudo codes are given for the above two algorithm processes,and specific driving scene experiments are constructed through numerical simulation and physical platform simulation(Pre Scan/Simulink co-simulation)respectively.The effectiveness of target estimation results of the proposed algorithms is judged by the relevant performance evaluation indicators,which provides theoretical support for the future application of the algorithms in actual driving scenarios. |
