Improving Human Action Recognition by Non-action Classification

TL;DR

This paper introduces a method to improve human action recognition in videos by identifying and down-weighting non-action segments using a trained non-action classifier, leading to more accurate recognition.

Contribution

The paper proposes a novel non-action classifier trained on shot-level annotated data to enhance action recognition accuracy by filtering irrelevant video segments.

Findings

Non-action segments can be effectively identified with high precision.

Using non-action classification improves overall action recognition performance.

The approach leverages the ActionThread dataset for training the classifier.

Abstract

In this paper we consider the task of recognizing human actions in realistic video where human actions are dominated by irrelevant factors. We first study the benefits of removing non-action video segments, which are the ones that do not portray any human action. We then learn a non-action classifier and use it to down-weight irrelevant video segments. The non-action classifier is trained using ActionThread, a dataset with shot-level annotation for the occurrence or absence of a human action. The non-action classifier can be used to identify non-action shots with high precision and subsequently used to improve the performance of action recognition systems.