Graph-Based Tracking with Uncertain ID Measurement Associations

Abstract

While multiple-hypothesis tracking is a leading paradigm for multi-sensor multi-target tracking, it is not effective in settings with disparate sensors that include high-rate kinematic data and low-rate identity data. Recent work has led to an effective graph-based approach to this challenge. This paper introduces two further advances: a generalization that allows for multiple (indistinguishable) objects of each type, and a scalable, time-based framework for hypothesis resolution. We illustrate promising performance results for multi-target track maintenance scenarios.