adityagirishpawateadityagirishpawate
12

Time Series Classification Using 1 D CNNs

screenshot of Time Series Classification Using 1 D CNNs

This project is on how to Develop 1D Convolutional Neural Network Models for Human Activity Recognition Below is an example video of a subject performing the activities while their movement data is being recorded. The six activities performed were as follows: Walking Walking Upstairs Walking D...

Github

Overview

The development of 1D Convolutional Neural Network (CNN) models for human activity recognition represents a significant advancement in the realm of machine learning applications. This innovative approach leverages motion data collected through smartphone sensors to accurately identify different physical activities. By processing accelerometer and gyroscopic data, the model is able to distinguish between activities such as walking, sitting, and standing, based on the movements recorded.

The remarkable aspect of this project lies in its real-world application, utilizing data collected from a Samsung Galaxy S II smartphone at a frequency of 50 Hz. Participants engaged in multiple activity sequences allows for a comprehensive dataset, paving the way for effective analysis and training of the CNN model.

Features

  • Human Activity Recognition: The model focuses on recognizing six different activities, providing flexibility for various application scenarios.

  • Comprehensive Data Collection: Utilizes accelerometer and gyroscopic data to capture not only linear movements but also angular velocity for improved accuracy.

  • High Sampling Rate: Data recorded at 50 Hz results in a rich dataset, allowing for detailed analysis and better learning outcomes.

  • Smartphone Integration: Conducted using a readily available Samsung Galaxy S II, making the technology accessible and practical for everyday use.

  • Strategic Data Pre-processing: Implements noise filters to enhance data quality, which is crucial for the machine learning process.

  • Effective Windowing Technique: Employs fixed windows of 2.56 seconds with 50% overlap, ensuring thorough coverage of activities for better algorithm performance.

  • Data Component Separation: Distinguishes between gravitational and motion components within the accelerometer data, optimizing the model's ability to learn relevant features.