You are on page 1of 1

Our Project steps for training poses of yoga/exercise using images data with labels of each pose for

supervised learning and testing on video frames and predict poses.

1- Data Gathering and Preparation

Stage Action Image


Collect images of yoga/exercise poses from various
Labeled Data Gathering sources.
Resize, augment, and normalize images for
Preprocessing consistency.

2- Keypoint Identification: Body Joints: Elbows, knees, wrists, etc., relevant to the pose.
3- Model Training and Optimization

Model Selection Architecture: CNN (PoseNet etc.)


Model Training:
Train/Validation/Test Sets: Dividing dataset for training, monitoring progress, and final
evaluation.
Hyperparameter Tuning: Adjustment of model parameters for optimal performance.
4- Video Pose Frame Extraction
Frame Extraction:
Divide Video: Split the video into individual frames from webcam.
Object Detection(Focus on Subject): Using any background subtraction or
motion detection algorithm to isolate the ROI.
5- Pose Prediction and Analysis:
Model Inference: Open camera to take video frames. Pass each frame through the
trained model to predict the pose and display alongside of the pose a person is standing, doing
pushup etc. or the trained poses.

You might also like