Two-Pass Zero-Shot Temporal-Spatial Grounding of Rare Traffic Events in Surveillance Video

Jiantang Huang · 2026

Read Two-Pass Zero-Shot Temporal-Spatial Grounding of Rare Traffic Events in Surveillance Video by Jiantang Huang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-01512.txt.

Description

Jiantang Huang locates rare traffic crashes in real security footage without any training on labeled crash video, a constraint set by the rules of the task. The system must name when a crash happens, where in the frame it sits, and what kind of collision it is, all at once.

The pipeline works in two passes. A first sweep at one frame per second gives a coarse time, place, and type. A second, denser pass inside a six-second window refines the call, with two fixed safety gates that fall back to the coarse answer when the fine pass hedges at edges. Two frozen vision-language models split the work, one grounding and one typing. On a benchmark of over two thousand real CCTV clips, the method beats the best baseline blend by a wide margin, at a total run cost of about twenty dollars.

Catalog record

CREATORJiantang Huang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.01512

Browse the full library · About the Tower of Babel