Seg2Track++: Probabilistic Track Validation and Data Association for Multi-Object Tracking and Segmentation
Researchers have developed Seg2Track++, a new framework for multi-object tracking and segmentation (MOTS) that enhances temporal consistency and identity preservation. The system integrates instance segmentation from SAM2 with a novel track management module. It uses Mask Centroid Distance and Confidence-Aware Cost Modulation for track association, and a Bernoulli filter for Probabilistic Track Validation to suppress false positives. Experiments on the KITTI MOTS dataset show improved performance without requiring fine-tuning. AI
IMPACT Enhances autonomous system reliability by improving object identity preservation and segmentation accuracy.