Research On Cooperative Multi-object Tracking Based On Probability Hypothesis Density Filter

With advanced perception technology,decision-making control technology,intelligent connected vehicle(ICV)has more and more functions of advanced driver assistant systems.It can not only improve vehicle safety and reduce road traffic accidents,but also further improve traffic efficiency and alleviate road congestion.In addition,it can also achieve eco-driving,so as to reduce energy consumption and environmental pollution.It is an important development direction in the future automotive field.The perception technology of ICV is the important foundation of decision-making control technology.High-precision and no blind spot perception will further improve vehicle safety.The traditional multi-object tracking technology of singlevehicle is limited by installation position and perceived range of the sensor,which has the problems of low tracking accuracy and large driving blind spot.With the rapid development of communication technology,inter-vehicle communication has become a reality.The information fusion of on-board sensors and inter-vehicle communication is an effective way to improve the perception accuracy and expand the perception range of ICV.Focusing on the cooperative multiobject tracking scene with known and unknown location data,a systematic study is carried out.The main work is as follows:(1)The probability hypothesis density filtering algorithm can’t adaptively generate the target and directly obtain the target state in the global coordinate system.Therefore,an adaptive global state estimation algorithm is proposed.Firstly,assuming that the number of clutter or false detection obeys Poisson distribution,the Poisson point process model is used to achieve the target adaptive birth.Then,the target state perceived by the intelligent vehicle and the positioning information obtained through the high-precision location system are integrated into the observation equation and the observation matrix.The experimental results show that this method not only solves the uncertainty of the appearance and disappearance of the target in the actual traffic scene,but also directly obtain the state estimation of the target in the global coordinate system.(2)Focusing on the problems of low perception accuracy and limited perception range of single vehicle multi-object tracking,a collaborative multi-object tracking framework is proposed based on the known location data.Firstly,the host vehicle and cooperative vehicle uses lidar to perceive the position of the target vehicle.Secondly,the state of the target in the global coordinate system is obtained through the above state estimation algorithm.Thirdly,the cooperative vehicle sends the target state to the host vehicle through inter-vehicle communication.Finally,the cooperative multi-object tracking is completed by using data association and data fusion algorithm.The simulation results based on motion model and Pre Scan show that compared with single vehicle multi-object tracking algorithm,the cooperative multi-object tracking method not only improves the perception accuracy,but also reduces the driving blind spot and expands the vehicle’s perception range.(3)The precise self-location of host vehicle and cooperative vehicle can significantly improve the performance of cooperative multi-object tracking.However,satellite navigation and positioning system is susceptible to the occlusion of high-rise buildings and tunnels,which makes it impossible to provide accurate location information.To this end,a track matching algorithm based on Bayesian inference is proposed.Firstly,the Bayesian probability model of joint trajectory association and relative pose estimation is established,and then the expectation maximization algorithm is used to iteratively estimate the relative pose and complete the data association.Simulation results show that the algorithm can accurately estimate the relative distance and heading of the host vehicle and cooperative vehicle,and can correctly correlate the observations of the same target.(4)A robust cooperative multi-object tracking framework is proposed based on the situation of unknown self-location information.The framework consists of three phases.Firstly,each vehicle perceives its surrounding environment based on the on-board sensors and exchanges the local tracks through inter-vehicle communication.On this basis,the above-mentioned track matching algorithm based on Bayesian inference is proposed,which realizes the track matching between the host vehicle and cooperative vehicle,and simultaneously optimizes the relative pose.Finally,a fast covariance crossover algorithm based on information theory is developed to fuse the tracks associated with the same target.A large number of simulation experiments show that this method successfully achieves cooperative multi-object tracking without the assistance of accurate self-location data,and improves the effect of multi-object tracking.