Human Pose estimation using Advance deep Neural Network

Abstract

This thesis is the culmination of my research in the rapidly advancing field of computer vision, with a particular focus on human pose estimation. Over the past few years, the importance of accurately and efficiently estimating human poses has grown significantly, driven by its application in various domains such as healthcare, sports analytics, and surveillance. My motivation for this research stems from the challenges and limitations observed in existing pose estimation techniques, particularly in complex scenarios involving dynamic movements and multiple interacting subjects. The primary objective of this research was to develop a robust and versatile pose estimation model capable of accurately predicting human poses in both 2D and 3D spaces. The research sought to address the limitations of traditional methods by leveraging the strengths of advanced neural network architectures. Specifically, the integration of ResNet-50 and Darknet-53 Convolutional Neural Networks (CNNs) with Bidirectional Long Short-Term Memory (BiLSTM) networks formed the backbone of the proposed hybrid models. These models were designed to capture both spatial and temporal information, thereby improving the accuracy and robustness of pose estimation in real-world applications. newline