##3D Pose Estimation from Video using CNN and RNN

This repository contains code for 3D pose estimation from video data using a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The project leverages MediaPipe for 2D keypoint extraction, TensorFlow/Keras for model building, and Hyperopt for hyperparameter optimization.

##Output of predicted 3d keypoints

output_3d_keypoints.mp4

##Output of predicted movements

output_predictions.mp4

Project Overview

##The goal of this project is to estimate 3D human poses from 2D video data. The pipeline includes:

2D Keypoint Extraction: Using MediaPipe to extract 2D keypoints from video frames. CNN-based 3D Pose Prediction: A CNN model is trained to predict 3D keypoints from the 2D keypoints. RNN-based Temporal Smoothing: An RNN model is used to precit future movements and smooth the 3D pose predictions across frames. Hyperparameter Optimization: Hyperopt is used to find the optimal model parameters. Combined Video Creation: A side-by-side video comparing the original footage and the 2D keypoints visualization.