基于SORT框架无迹卡尔曼滤波的多目标跟踪算法

Multi-Target Tracking Algorithm Based on Unscented Kalman Filter within SORT Framework

摘要:

针对同向密集目标跟踪需求，提出了一种基于SORT（simple online and realtime tracking）框架无迹卡尔曼滤波（unscented Kalman filter，UKF）的多目标跟踪算法。该算法利用SORT框架构建多无迹卡尔曼滤波（multiple unscented Kalman filter，MUKF）跟踪器，使用欧氏距离和马氏距离计算代价矩阵，采用匈牙利算法和贪婪算法进行数据关联匹配，并根据匹配结果进行标签管理以及生命周期管理，实现多目标跟踪。仿真结果表明：与卡尔曼滤波（Kalman filter，KF）算法、扩展卡尔曼滤波（extended Kalman filter，EKF）算法相比，所提算法也具有良好的多目标跟踪性能，且基于马氏距离的多目标跟踪性能更优。

Abstract:

A multi-target tracking algorithm based on unscented Kalman filter (UKF) within simple online and realtime tracking (SORT) framework was proposed to meet the requirement of intensive target tracking in the same direction. This algorithm employed the SORT framework to construct a multiple unscented Kalman filter (MUKF) tracker. The cost matrix was calculated using Euclidean distance and Mahalanobis distance, the Hungarian algorithm and the greedy algorithm were applied for data association matching, and label management and life cycle management were performed on the basis of the matching results to achieve multi-target tracking. Simulation results demonstrate that compared with Kalman filter (KF) algorithm and extended Kalman filter (EKF) algorithm, the proposed algorithm has good performance in multi-target tracking, and the performance is better when based on Mahalanobis distance.