Data Processing Pipeline

A human activity recognition data processing pipeline dedicated at transforming triaxial smartphone data into a 580+ feature vector, usable by classifiers for Human Activity Recognition.

Overview

The HAR-Processing-Pipeline is dedicated to the processing of triaxial smartphone data for HAR. It provides a comprehensive solution for data augmentation and dataset orchestration, aiming to facilitate the transformation of raw sensor data into a feature-rich format suitable for HAR classification.

Features

• Triaxial smartphone data processing
• Feature vector generation
• Dataset orchestration for HAR
• End-to-end pipeline for HAR augmentation

Details

Dataset

An android application was developed to ease the process of collecting and labelling movement data. 7 entries at 50hz were taken namely:

• Triaxial Linear Acceleration
• Triaxial Angular Acceleration
• Time.

Data proportions

Sliding Window Technique

Performance

GNB = Gaussian Naive Bayes SGD = Stochastic Gradient Descent SVM = Support Vector Machine

OD = Our Data

SVM:

CNN