DeepSegmenter: Temporal Action Localization for Detecting Anomalies in Untrimmed Naturalistic Driving Videos

TL;DR

DeepSegmenter is a novel framework that simultaneously segments and classifies activities in untrimmed naturalistic driving videos to detect anomalies, improving understanding of driver behavior.

Contribution

It introduces a unified framework combining activity segmentation and classification for continuous driving videos, addressing limitations of previous classification-only approaches.

Findings

Achieved 8th place in the 2023 AI City Challenge, Track 3.

Attained an activity overlap score of 0.5426 on validation data.

Demonstrated effectiveness, efficiency, and robustness of the method.

Abstract

Identifying unusual driving behaviors exhibited by drivers during driving is essential for understanding driver behavior and the underlying causes of crashes. Previous studies have primarily approached this problem as a classification task, assuming that naturalistic driving videos come discretized. However, both activity segmentation and classification are required for this task due to the continuous nature of naturalistic driving videos. The current study therefore departs from conventional approaches and introduces a novel methodological framework, DeepSegmenter, that simultaneously performs activity segmentation and classification in a single framework. The proposed framework consists of four major modules namely Data Module, Activity Segmentation Module, Classification Module and Postprocessing Module. Our proposed method won 8th place in the 2023 AI City Challenge, Track 3, with an activity overlap score of 0.5426 on experimental validation data. The experimental results demonstrate the effectiveness, efficiency, and robustness of the proposed system.