Analytics module for correction and feedback of long jump athlete using anomaly detection

Pradyumn Vikram

Usage

Install required dependancies by running
pip install mediapipe opencv-python scikit-learn numpy
Run the module via
python -m analytics_module.py

Repository Structure

• The main executable for this projecy is analytics_module.py, which takes a trimmed long jump sequence as input, giving corrective feedback as output
• generator_trimmer.ipynb contains the pipeline for extracting the dataset and processing it for phase segmentation - it requries the data directory with altered_codec, final_dataset, processed, raw_data, ref_vid, split_videos sub directories - these have not been included in the repo since the content is downloaded by the notebook automatically (Note: Convert the downloaded videos (saved in raw_data) to a suitable codec for processing and transfer them to the altered_codec directory)
• preprocessing_pose_extraction.ipynb contains the pipeline for extracting clusters and segmenting phases
• train_extracted_features_anomaly_detection.ipynb contains the pipeline for training the final anomaly detection model
• All models are saved in models/ directory and the scalers are saved in scalers/
• project report and test video can be found in the misc folder
• keypoints contains extracted feature arrays which are processed furthur for anomaly detection

Project Report

A report for the project can be found here