Online Detection of Action Start in Untrimmed, Streaming Videos

TL;DR

This paper introduces the task of online action start detection in untrimmed streaming videos, proposing three novel methods to improve detection accuracy and latency, validated through extensive experiments on benchmark datasets.

Contribution

It presents three innovative techniques—GAN-based hard negative sample generation, temporal consistency modeling, and adaptive sampling—for training ODAS models, advancing the state-of-the-art.

Findings

Significant performance improvements over existing methods.

Effective handling of ambiguous backgrounds and data scarcity.

Validated on THUMOS'14 and ActivityNet datasets.

Abstract

We aim to tackle a novel task in action detection - Online Detection of Action Start (ODAS) in untrimmed, streaming videos. The goal of ODAS is to detect the start of an action instance, with high categorization accuracy and low detection latency. ODAS is important in many applications such as early alert generation to allow timely security or emergency response. We propose three novel methods to specifically address the challenges in training ODAS models: (1) hard negative samples generation based on Generative Adversarial Network (GAN) to distinguish ambiguous background, (2) explicitly modeling the temporal consistency between data around action start and data succeeding action start, and (3) adaptive sampling strategy to handle the scarcity of training data. We conduct extensive experiments using THUMOS'14 and ActivityNet. We show that our proposed methods lead to significant performance gains and improve the state-of-the-art methods. An ablation study confirms the effectiveness of each proposed method.