A Classification-Aided Recurrent Neural Network for Tracking in Urban Terrain

Conference Paper

This paper addresses the challenge of tracking small unmanned aerial systems in urban terrain. Tracking is made up of two parts, state estimation and data association. For low, slow, small targets in the presence of other non-Target movers with similar dynamics, data association is arguably the greatest challenge. In traditional trackers, measurements are associated with existing tracks based on hand-crafted dynamical models. In this paper, we present recurrent neural network (RNN) based tracker which uses both learned target dynamic models and classification scores to associate measurements with target tracks. Classification scores for each radar measurement are generated by a neural-network trained using in situ 'pattern-of-life' radar data. These scores together with location estimates are then passed to RNN trackers with long-short-Term-memory (LSTM) trained for each object class. In this paper, we present simulated and real data results indicating that classification scores significantly improves small UAS track holding times in urban environments.

Full Text

Duke Authors

Cited Authors

  • Martinez, MA; Krolik, JL

Published Date

  • January 1, 2022

Published In

  • Proceedings of the Ieee Radar Conference

Electronic International Standard Serial Number (EISSN)

  • 2375-5318

International Standard Serial Number (ISSN)

  • 1097-5764

Digital Object Identifier (DOI)

  • 10.1109/RadarConf2248738.2022.9764254

Citation Source

  • Scopus