Human Activity Recognition via Spectrotemporal Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional patient data acquisition methods are invasive or require manual entry, limiting their pervasiveness in daily life, and thus fail to effectively monitor chronic diseases such as diabetes and obesity, which are expected to rise significantly by 2030.
Innovation Solution
A system that uses deep learning models, specifically 2D-CNN and 1D-CNN, to recognize human activities by transforming raw sensor data into spectrotemporal representation and classifying activities, providing a non-invasive and unobtrusive method for continuous monitoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional patient data acquisition methods (invasive sensors or manual entry) are used, then data collection can be performed, but patient tolerance and pervasiveness in daily life are reduced
Solution Approach 1:
The patent replaces invasive mechanical sensors with non-invasive optical sensors that detect movements through light reflection off the skin. This substitution eliminates the need for uncomfortable mechanical contact while maintaining continuous activity monitoring capability, thereby improving patient tolerance and pervasiveness in daily life
Solution Approach 2:
The system automatically captures images and processes them through machine learning models without requiring manual intervention. The smartphone camera continuously captures images, and the system autonomously analyzes movements to detect activities, eliminating the need for patients to manually enter data while maintaining reliable monitoring
2Measurement precision
If deep learning models with spectrotemporal representation are used, then classification accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational task by using pre-trained machine learning models that process simplified spectrotemporal representations rather than raw video data. This segmentation allows the system to maintain high classification accuracy while reducing the computational burden on mobile devices, as the complex model processing is performed on prepared feature data rather than raw images
Solution Approach 2:
The system transforms raw image data into spectrotemporal representation data through parameter transformations (frequency domain conversion). This transformation changes the data parameters to a format that is more suitable for machine learning classification, improving accuracy while the transformation itself is computationally efficient compared to processing raw video streams
Data Source
AI summary
Methods and systems for network scanning activity detection are disclosed. Various embodiments of the methods and systems may include: obtaining raw sensor data for a user; generating spectrotemporal representation data corresponding to the raw sensor data; applying the spectrotemporal representation data to a trained deep learning model; receiving a classification indication from the trained deep learning model; and providing a human activity indication of the user based on the classification indication. Other aspects, embodiments, and features are also claimed and described.