Enhanced Noise-Resilient Pressure Mat System Based on Hyperdimensional Computing.

Area of Science:

• Sensor Technology
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Traditional indoor pressure sensing and human activity recognition (HAR) systems often utilize expensive, high-resolution mats and computationally demanding neural network (NN) models.
• These conventional approaches are susceptible to noise, impacting their reliability and efficiency for real-world applications.

Purpose of the Study:

• To design and evaluate a cost-effective and noise-resilient pressure mat system for HAR.
• To leverage intelligent pressure sensing with Velostat and a novel hyperdimensional computing (HDC) classifier for improved performance.
• To introduce a new HDC algorithm specifically designed for noise resilience.

Main Methods:

• Development of a pressure mat system using Velostat for intelligent pressure sensing.
• Implementation of a novel hyperdimensional computing (HDC) classifier for HAR.
• Collection of two datasets to capture static and continuous human movements.
• Analysis of the system's performance under various noise conditions (memory/communication, input, sensor, Gaussian).

Main Results:

• The HDC-based classification algorithm achieved an accuracy of 93.19%, outperforming state-of-the-art Convolutional Neural Networks (CNNs) by 9.47%.
• The system demonstrated an 85% reduction in energy consumption compared to traditional methods.
• The proposed HDC algorithm exhibited enhanced resilience to memory/communication, input, and sensor noise.
• In the presence of Gaussian noise, the system achieved 92.15% accuracy (97.51% for static data), a significant improvement over CNNs.

Conclusions:

• The proposed cost-effective pressure mat system with HDC offers a superior alternative for HAR, providing higher accuracy and robustness against noise.
• The novel HDC noise-resilient algorithm significantly enhances system reliability in challenging environments.
• This approach presents a promising direction for low-power, high-performance HAR systems.