A Survey on Artificial Intelligence in Posture Recognition.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Technological advancements have significantly expanded the scope and applications of posture recognition.
• Research in posture recognition has evolved with new methodologies and algorithms.

Purpose of the Study:

• To introduce the latest methods in posture recognition.
• To review recent techniques and algorithms, including CNNs and their enhancements.
• To analyze datasets and compare different recognition approaches.

Main Methods:

• Review of traditional methods: Scale-Invariant Feature Transform (SIFT), Histogram of Oriented Gradients (HOG), Support Vector Machine (SVM), Gaussian Mixture Model (GMM), Dynamic Time Warping (DTW), Hidden Markov Model (HMM).
• Investigation of Convolutional Neural Network (CNN) advancements: Stacked Hourglass Networks, Multi-stage Pose Estimation Networks, Convolutional Pose Machines, High-Resolution Nets.
• Exploration of advanced neural network applications: Transfer Learning, Ensemble Learning, Graph Neural Networks, Explainable Deep Neural Networks.

Main Results:

• CNNs have demonstrated significant success and are favored by researchers in posture recognition.
• Hidden Markov Models (HMM) and Support Vector Machines (SVM) remain widely used classification methods.
• Lightweight networks are emerging as a notable area of research interest.

Conclusions:

• While CNNs excel, further research is needed in feature extraction and information fusion for posture recognition.
• The development of 3D benchmark datasets and data generation techniques is a critical future research direction.
• Continued exploration of advanced neural network architectures is essential for advancing the field.