Human Activity Recognition Using Pose Estimation, Deep Learning, And Contextual Scene Analysis
Kandula Vamshi Krishna, T. Malathi · International Journal of Research Publication and Reviews · 2025
The human activity recognition (HAR) system, designed to classify activities, such as standing, sitting, walking, running, jumping, lying, lying, climbing stairs, and leaning from live camera feeds, stable images and videos.System 3D pose combines mediapipes, a long-term short-term memory (LSTM) model UCI trained on HAR dataset, and bootstraping language-image pre-training (BLIP) for relevant visual analysis.A PYQT5-based graphical user interface integrates these components to provide real-time recognition in several input mode.Experimental testing shows reliable performance at ~ 30 FPS, with accuracy accuracy to increase BLIP in vague scenarios.This multimodal system is versatile for healthcare monitoring, fitness tracking, monitoring and human -computer interactions.