Space-oriented Label Partitioning for Multi-object Tracking

The problem of tracking multiple objects has been investigated in various research and industrial fields. Among existing methods, random finite set (RFS) solutions such as the generalized labeled multi-Bernoulli (GLMB) filter has provided efficient solutions with solid theoretical justifications. Furthermore, implementations show that the GLMB approach is efficient under challenging scenarios. In this paper, we study an RFS-based method for multi-object tracking (MOT) through a simple data structure for label partitioning. Specifically, grid index structure based techniques for splitting a label space and a label-partitioned GLMB tracker are investigated. We finally evaluate the performance of label partitioning and the GLMB filter via various means such as visualization, execution time, and MOT metrics.