Transformer-Based Maneuvering Target Tracking

When tracking maneuvering targets, recurrent neural networks (RNNs), especially long short-term memory (LSTM) networks, are widely applied to sequentially capture the motion states of targets from observations. However, LSTMs can only extract features of trajectories stepwise; thus, their modeling o...

Full description

Bibliographic Details
Main Authors: Guanghui Zhao, Zelin Wang, Yixiong Huang, Huirong Zhang, Xiaojing Ma
Format: Article
Language:English
Published: MDPI AG 2022-11-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/22/21/8482
_version_ 1797466456270569472
author Guanghui Zhao
Zelin Wang
Yixiong Huang
Huirong Zhang
Xiaojing Ma
author_facet Guanghui Zhao
Zelin Wang
Yixiong Huang
Huirong Zhang
Xiaojing Ma
author_sort Guanghui Zhao
collection DOAJ
description When tracking maneuvering targets, recurrent neural networks (RNNs), especially long short-term memory (LSTM) networks, are widely applied to sequentially capture the motion states of targets from observations. However, LSTMs can only extract features of trajectories stepwise; thus, their modeling of maneuvering motion lacks globality. Meanwhile, trajectory datasets are often generated within a large, but fixed distance range. Therefore, the uncertainty of the initial position of targets increases the complexity of network training, and the fixed distance range reduces the generalization of the network to trajectories outside the dataset. In this study, we propose a transformer-based network (TBN) that consists of an encoder part (transformer layers) and a decoder part (one-dimensional convolutional layers), to track maneuvering targets. Assisted by the attention mechanism of the transformer network, the TBN can capture the long short-term dependencies of target states from a global perspective. Moreover, we propose a center–max normalization to reduce the complexity of TBN training and improve its generalization. The experimental results show that our proposed methods outperform the LSTM-based tracking network.
first_indexed 2024-03-09T18:40:08Z
format Article
id doaj.art-6feef1edbae142eaa7178ef68005f3ec
institution Directory Open Access Journal
issn 1424-8220
language English
last_indexed 2024-03-09T18:40:08Z
publishDate 2022-11-01
publisher MDPI AG
record_format Article
series Sensors
spelling doaj.art-6feef1edbae142eaa7178ef68005f3ec2023-11-24T06:48:46ZengMDPI AGSensors1424-82202022-11-012221848210.3390/s22218482Transformer-Based Maneuvering Target TrackingGuanghui Zhao0Zelin Wang1Yixiong Huang2Huirong Zhang3Xiaojing Ma4School of Artificial Intelligence, Xidian University, Xi’an 710071, ChinaSchool of Artificial Intelligence, Xidian University, Xi’an 710071, ChinaSchool of Artificial Intelligence, Xidian University, Xi’an 710071, ChinaSchool of Artificial Intelligence, Xidian University, Xi’an 710071, ChinaSchool of Electronic Confrontation, National University of Defense, Hefei 230037, ChinaWhen tracking maneuvering targets, recurrent neural networks (RNNs), especially long short-term memory (LSTM) networks, are widely applied to sequentially capture the motion states of targets from observations. However, LSTMs can only extract features of trajectories stepwise; thus, their modeling of maneuvering motion lacks globality. Meanwhile, trajectory datasets are often generated within a large, but fixed distance range. Therefore, the uncertainty of the initial position of targets increases the complexity of network training, and the fixed distance range reduces the generalization of the network to trajectories outside the dataset. In this study, we propose a transformer-based network (TBN) that consists of an encoder part (transformer layers) and a decoder part (one-dimensional convolutional layers), to track maneuvering targets. Assisted by the attention mechanism of the transformer network, the TBN can capture the long short-term dependencies of target states from a global perspective. Moreover, we propose a center–max normalization to reduce the complexity of TBN training and improve its generalization. The experimental results show that our proposed methods outperform the LSTM-based tracking network.https://www.mdpi.com/1424-8220/22/21/8482attention mechanismmaneuvering target trackingrecurrent neural networktransformer-based network
spellingShingle Guanghui Zhao
Zelin Wang
Yixiong Huang
Huirong Zhang
Xiaojing Ma
Transformer-Based Maneuvering Target Tracking
Sensors
attention mechanism
maneuvering target tracking
recurrent neural network
transformer-based network
title Transformer-Based Maneuvering Target Tracking
title_full Transformer-Based Maneuvering Target Tracking
title_fullStr Transformer-Based Maneuvering Target Tracking
title_full_unstemmed Transformer-Based Maneuvering Target Tracking
title_short Transformer-Based Maneuvering Target Tracking
title_sort transformer based maneuvering target tracking
topic attention mechanism
maneuvering target tracking
recurrent neural network
transformer-based network
url https://www.mdpi.com/1424-8220/22/21/8482
work_keys_str_mv AT guanghuizhao transformerbasedmaneuveringtargettracking
AT zelinwang transformerbasedmaneuveringtargettracking
AT yixionghuang transformerbasedmaneuveringtargettracking
AT huirongzhang transformerbasedmaneuveringtargettracking
AT xiaojingma transformerbasedmaneuveringtargettracking