DEEP LEARNING - BASED HUMAN POSE ESTIMATION METHODS FOR TRAINING ACTIVITIES: A COMPREHENSIVE REVIEW AND FRAMEWORK DESIGN FOR INTELLIGENT SHOOTING TRAINING SYSTEMS

Abstract

Human Pose Estimation (HPE) has witnessed significant advancements in recent years, largely propelled by
the breakthroughs in deep learning techniques. This paper presents a comprehensive review of Deep Learning-
Based HPE Methods in the context of trainning activities. This review begins by introducing the fundamental
concepts of human pose estimation and its significance in sports and physical training. Subsequently, the paper
delves into the landscape of deep learning methodologies, including convolutional neural networks (CNNs),
recurrent neural networks (RNNs), and their variants, which have played a pivotal role in revolutionizing pose
estimation. The core of the review lies in the analysis of state-of-the-art deep learning-based HPE methods
tailored to training activities. Emphasis is placed on their accuracy, robustness to varying conditions, real-time
processing capabilities, and integration with training environments. The challenges and limitations of deep
learning-based HPE methods are addressed, along with ongoing research efforts to mitigate these issues. Finally,
this paper proposes a shooting training model based on integrating of knowledge distillation into the context of
HPE, highlighting its potential to enhance training assistance systems.