Action Transformer: A self-attention model for short-time pose-based human action recognition. (April 2022)

• Record Type: Journal Article
• Authors: Mazzia, Vittorio; Angarano, Simone; Salvetti, Francesco; Angelini, Federico; Chiaberge, Marcello
• Is Part Of: Pattern recognition. Volume 124(2022)
• Publication Date: 2022-04
• Subjects: Human action recognition -- Deep learning -- Computer vision -- Transformer; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.108487
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We study the application of the Transformer encoder to 2D pose-based HAR and propose the novel AcT model. We introduce MPOSE2021, a dataset for real-time short-time HAR. In contrast to other publicly available datasets, the peculiarity of having a constrained number of time steps stimulates the development of actual real-time methodologies that perform HAR with low latency and high throughput. We conduct extensive experimentation on model performance and latency to verify the suitability of AcT for real-time applications. Abstract: Deep neural networks based purely on attention have been successful across several domains, relying on minimal architectural priors from the designer. In Human Action Recognition (HAR), attention mechanisms have been primarily adopted on top of standard convolutional or recurrent layers, improving the overall generalization capability. In this work, we introduce Action Transformer (AcT), a simple, fully, self-attentional architecture that consistently outperforms more elaborated networks that mix convolutional, recurrent, and attentive layers. In order to limit computational and energy requests, building on previous human action recognition research, the proposed approach exploits 2D pose representations over small temporal windows, providing a low latency solution for accurate and effective real-time performance. Moreover, we open-source MPOSE2021, a new large-scale dataset, as an attempt to build a formal training and evaluation benchmark for real-time, short-time HAR. The proposed methodology was extensively tested on MPOSE2021 and compared to several state-of-the-art architectures, proving the effectiveness of the AcT model and laying the foundations for future work on HAR.