Towards Robust Human Activity Recognition from RGB Video Stream with Limited Labeled Data

TL;DR

This paper introduces a novel RGB-only human activity recognition framework using skeleton data and deep BLSTM, overcoming limited labeled data challenges and outperforming RGB-D based methods.

Contribution

The paper presents a new RGB-only activity recognition approach with innovative training techniques that surpass state-of-the-art RGB-D methods.

Findings

RGB-only approach outperforms RGB-D methods

Effective training with limited labeled data

Proposed techniques improve model robustness

Abstract

Human activity recognition based on video streams has received numerous attentions in recent years. Due to lack of depth information, RGB video based activity recognition performs poorly compared to RGB-D video based solutions. On the other hand, acquiring depth information, inertia etc. is costly and requires special equipment, whereas RGB video streams are available in ordinary cameras. Hence, our goal is to investigate whether similar or even higher accuracy can be achieved with RGB-only modality. In this regard, we propose a novel framework that couples skeleton data extracted from RGB video and deep Bidirectional Long Short Term Memory (BLSTM) model for activity recognition. A big challenge of training such a deep network is the limited training data, and exploring RGB-only stream significantly exaggerates the difficulty. We therefore propose a set of algorithmic techniques to train this model effectively, e.g., data augmentation, dynamic frame dropout and gradient injection. The experiments demonstrate that our RGB-only solution surpasses the state-of-the-art approaches that all exploit RGB-D video streams by a notable margin. This makes our solution widely deployable with ordinary cameras.