Csongor Jánosi

Abstract

Real-time 2D human pose estimation using smartphone cameras has
become a highly relevant technology for interactive applications
such as physical education, gesture-based interaction, and aug-
mented reality. While recent advances have significantly improved
pose estimation models, using such systems on mobile devices
remains challenging due to computational, energy, and latency con-
straints. This paper presents a state-of-the-art review of real-time
2D pose estimation approaches targeting smartphones, following
a PRISMA inspired systematic literature review. Two recent and
relevant papers, LMFormer and LiPE, are analyzed and compared
based on design choices, efficiency metrics, evaluation scope, and
suitability for mobile and human–computer interaction (HCI) prac-
tical use cases. The review highlights the balancing of accuracy and
computational efficiency, identifies limitations in current evaluation
practices, and outlines open research gaps related to end-to-end
mobile performance and interaction-oriented evaluation criteria.