TY - JOUR
T1 - LAAN-ANMFF
T2 - An adaptive data-driven association filter for multiple target tracking in clutter
AU - Yu, Mohan
AU - Gao, Meiguo
N1 - Publisher Copyright:
© 2026 Elsevier Inc.
PY - 2026/11/15
Y1 - 2026/11/15
N2 - Robust multiple target tracking (MTT) in cluttered environments remains challenging due to severe measurement ambiguity and association uncertainty. Existing methods, which rely on limited measurement information and continuous trajectories, suffer from degraded association performance under such conditions. A lightweight association assessment network based adaptive noise-aware multi-measurement fusion filter (LAAN-ANMFF) is proposed to enhance association reliability and state estimation. The network estimates association probabilities from multi-feature inputs, and incorporates a competition-aware modulation to suppress ambiguous associations for MTT. To better exploit sufficient measurement data, a multi-measurement fusion framework within Kalman filter is adopted, where association probabilities are further used to enable adaptive noise-aware and entropy-guided uncertainty modeling during fusion. An equivalent fused noise covariance is then derived via maximum likelihood estimation to characterize the overall uncertainty. Simulation experiments are conducted under challenging conditions with high-density and strong-interference clutter. Results demonstrate that the proposed algorithm achieves superior tracking accuracy, stability and adaptability compared with existing methods, while maintaining low computational complexity.
AB - Robust multiple target tracking (MTT) in cluttered environments remains challenging due to severe measurement ambiguity and association uncertainty. Existing methods, which rely on limited measurement information and continuous trajectories, suffer from degraded association performance under such conditions. A lightweight association assessment network based adaptive noise-aware multi-measurement fusion filter (LAAN-ANMFF) is proposed to enhance association reliability and state estimation. The network estimates association probabilities from multi-feature inputs, and incorporates a competition-aware modulation to suppress ambiguous associations for MTT. To better exploit sufficient measurement data, a multi-measurement fusion framework within Kalman filter is adopted, where association probabilities are further used to enable adaptive noise-aware and entropy-guided uncertainty modeling during fusion. An equivalent fused noise covariance is then derived via maximum likelihood estimation to characterize the overall uncertainty. Simulation experiments are conducted under challenging conditions with high-density and strong-interference clutter. Results demonstrate that the proposed algorithm achieves superior tracking accuracy, stability and adaptability compared with existing methods, while maintaining low computational complexity.
KW - Adaptive noise estimation
KW - Cluttered environments
KW - Data association
KW - Deep learning
KW - Kalman filter
KW - Multiple target tracking
UR - https://www.scopus.com/pages/publications/105051221464
U2 - 10.1016/j.dsp.2026.106499
DO - 10.1016/j.dsp.2026.106499
M3 - Article
AN - SCOPUS:105051221464
SN - 1051-2004
VL - 184
JO - Digital Signal Processing: A Review Journal
JF - Digital Signal Processing: A Review Journal
M1 - 106499
ER -