CANMOT replaces global noise covariances in KF-based 3D MOT with class-specific diagonal matrices (optionally object-frame) and shows gains in accuracy and ID-switch reduction on nuScenes plus improved but still inconsistent uncertainty calibration.
Weak in the NEES?: Auto-tuning kalman filters with bayesian optimization,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CANMOT: Class-Aware Noise Modeling for Multi-Object Tracking in Autonomous Driving
CANMOT replaces global noise covariances in KF-based 3D MOT with class-specific diagonal matrices (optionally object-frame) and shows gains in accuracy and ID-switch reduction on nuScenes plus improved but still inconsistent uncertainty calibration.