一种未知环境下机器人多目标跟踪算法

A robot multi-object tracking algorithm in unknown environments

针对未知环境下移动机器人多目标跟踪问题,设计了一种基于联合概率数据关联的粒子滤波算法。该算法利用联合概率数据关联方法对Rao-Blackwellized粒子滤波算法进行改进,使机器人能够完成未知环境条件下对自身状态、环境特征状态和多目标状态的在线联合估计。算法将系统状态变量分为代表多目标、环境特征状态的线性变量和代表机器人状态的非线性变量,并利用联合概率数据关联Kalman滤波和粒子滤波对系统状态进行更新。通过仿真实验证明了该算法对机器人状态、环境特征状态以及多目标状态的估计准确性,验证了算法对未知环境下多目标的跟踪能力。

In this paper,a particle filtering algorithm based on the joint integrated probabilistic data association（ JIPDA） is proposed in order to solve the problem of motile robot multi-object tracking in unknown environments.The Rao-Blackwellized particle filtering is reconstructed based on the JIPDA in the new algorithm. It allows the robot to estimate joint states of itself,environment features and multi-object states simultaneously. The algorithm divides the system variables into two parts： the lineal variable representing multi-object and environment feature states,and the non-linear variable representing robot states. The system state is updated by JIPDA Kalman filtering and particle filtering. Estimation precision of robot states,environment feature states and multi-object states is verified by simulation results,verifying the ability of multi-object tracking in unknown environments.