TRAQ-query: trajectory reliability-aware query modulation for transformer-based multi-object tracking

Transformer-based multi-object tracking (MOT) propagates track queries across frames, but missed observations, partial occlusion, camera motion, and ambiguous interactions can degrade these temporal states. We propose TRAQ-query, a lightweight trajectory reliability-aware query modulation method for MOTR-style online tracking. TRAQ-query estimates the propagation reliability of each active track query from its latent representation and short-term box-state variation, using a motion-consistency intersection-over-union (MCIoU) soft target during adaptation. After the standard query-interaction update, the predicted reliability controls a bounded interpolation between the updated query and a fresh-query reference. The backbone, transformer, prediction heads, query-interaction module, and matching procedure remain frozen, so only the lightweight reliability head is optimized. On the complete 200-video BDD100K vehicle validation split under the controlled proposal setting, TRAQ-query improves Higher Order Tracking Accuracy (HOTA) from 16.657 to 18.252, Multiple Object Tracking Accuracy (MOTA) from 1.769 to 2.333, and Identification F1 score (IDF1) from 12.646 to 14.562, while reducing false negatives by 11,780. We further evaluate TRAQ-query on DanceTrack using the official pretrained MOTRv2 checkpoint and proposal resources. Relative to the published MOTRv2 reference, which reports 69.9 HOTA, 91.9 MOTA, and 71.7 IDF1, TRAQ-query achieves 70.7 HOTA, 92.2 MOTA, and 72.9 IDF1. These complementary results indicate that reliability-conditioned query modulation can benefit temporal query propagation across distinct evaluation settings.

Authors

Institutions

Publication Details

Journal
Journal of King Saud University - Computer and Information Sciences
Published
2026-11-18
DOI
https://doi.org/10.1007/s44443-026-01495-8
Primary Topic
Video Surveillance and Tracking Methods
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

TRAQ-query: trajectory reliability-aware query modulation for transformer-based multi-object tracking

Wenjing Gao
2 links
Journal of King Saud University - Computer and Information Sciences
Video Surveillance and Tracking Methods
article

TRAQ-query: trajectory reliability-aware query modulation for transformer-based multi-object tracking

Wenjing Gao
article en

Abstract

Transformer-based multi-object tracking (MOT) propagates track queries across frames, but missed observations, partial occlusion, camera motion, and ambiguous interactions can degrade these temporal states. We propose TRAQ-query, a lightweight trajectory reliability-aware query modulation method for MOTR-style online tracking. TRAQ-query estimates the propagation reliability of each active track query from its latent representation and short-term box-state variation, using a motion-consistency intersection-over-union (MCIoU) soft target during adaptation. After the standard query-interaction update, the predicted reliability controls a bounded interpolation between the updated query and a fresh-query reference. The backbone, transformer, prediction heads, query-interaction module, and matching procedure remain frozen, so only the lightweight reliability head is optimized. On the complete 200-video BDD100K vehicle validation split under the controlled proposal setting, TRAQ-query improves Higher Order Tracking Accuracy (HOTA) from 16.657 to 18.252, Multiple Object Tracking Accuracy (MOTA) from 1.769 to 2.333, and Identification F1 score (IDF1) from 12.646 to 14.562, while reducing false negatives by 11,780. We further evaluate TRAQ-query on DanceTrack using the official pretrained MOTRv2 checkpoint and proposal resources. Relative to the published MOTRv2 reference, which reports 69.9 HOTA, 91.9 MOTA, and 71.7 IDF1, TRAQ-query achieves 70.7 HOTA, 92.2 MOTA, and 72.9 IDF1. These complementary results indicate that reliability-conditioned query modulation can benefit temporal query propagation across distinct evaluation settings.

Journal of King Saud University - Computer and Information SciencesVol. 38(10)
Shandong University (CN)
Video Surveillance and Tracking Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon