TrackTeller: Temporal Multimodal 3D Grounding for Behavior-Dependent Object References

TL;DR

TrackTeller introduces a temporal multimodal framework for 3D object grounding in dynamic driving scenes, leveraging multi-frame data and language cues to improve accuracy and reduce false alarms.

Contribution

It presents a novel unified architecture that combines LiDAR-image fusion, language-conditioned decoding, and temporal reasoning for behavior-dependent object referencing.

Findings

Achieves 70% improvement in tracking accuracy on NuPrompt benchmark.

Reduces false alarm frequency by over 3 times.

Outperforms existing baselines in dynamic 3D grounding tasks.

Abstract

Understanding natural-language references to objects in dynamic 3D driving scenes is essential for interactive autonomous systems. In practice, many referring expressions describe targets through recent motion or short-term interactions, which cannot be resolved from static appearance or geometry alone. We study temporal language-based 3D grounding, where the objective is to identify the referred object in the current frame by leveraging multi-frame observations. We propose TrackTeller, a temporal multimodal grounding framework that integrates LiDAR-image fusion, language-conditioned decoding, and temporal reasoning in a unified architecture. TrackTeller constructs a shared UniScene representation aligned with textual semantics, generates language-aware 3D proposals, and refines grounding decisions using motion history and short-term dynamics. Experiments on the NuPrompt benchmark demonstrate that TrackTeller consistently improves language-grounded tracking performance, outperforming strong baselines with a 70% relative improvement in Average Multi-Object Tracking Accuracy and a 3.15-3.4 times reduction in False Alarm Frequency.