Stats-Calculus Pose Descriptor Feeding A Discrete HMM Low-latency Detection and Recognition System For 3D Skeletal Actions

TL;DR

This paper introduces a low-latency human action recognition system using a novel set of pose descriptors based on Stats-Calculus features and Hidden Markov Models, improving real-time 3D skeletal action detection.

Contribution

It presents a new statistical pose descriptor set and integrates them with HMMs for efficient low-latency 3D skeletal action recognition, which is a novel approach in this domain.

Findings

Effective in low-latency action detection

Improved recognition accuracy over traditional methods

Demonstrates robustness with various pose descriptors

Abstract

Recognition of human actions, under low observational latency, is a growing interest topic, nowadays. Many approaches have been represented based on a provided set of 3D Cartesian coordinates system originated at a certain specific point located on a root joint. In this paper, We will present a statistical detection and recognition system using Hidden Markov Model using 7 types of pose descriptors. * Cartesian Calculus Pose descriptor. * Angular Calculus Pose descriptor. * Mixed-mode Stats-Calculus Pose descriptor. * Centro-Stats-Calculus Pose descriptor. * Rela-Centro-Stats-Calculus Pose descriptor. * Rela-Centro-Stats-Calculus DCT Pose descriptor. * Rela-Centro-Stats-Calculus DCT-AMDF Pose descriptor. Stats-Calculus is a feature extracting technique, that is developed on Moving Pose descriptor , but using a combination of Statistics measures and Calculus measures.